> For Diffio's key facts and every Markdown page, see [llms.txt](https://diffio.ai/llms.txt). This page is also available as HTML at https://diffio.ai/press/.

Diffio AI press kit: before and after speech restoration examples in audio and video, no-logo editorial cuts, screenshots, logos and a fact sheet.

# Diffio press kit

Download before and after audio and video of recorded speech, spectrograms, product screenshots and logos, or use the boilerplate and fact sheet below. Everything here is free to use in coverage.

Press contact

Nathan Harmon, [nathan.harmon@diffio.ai](mailto:nathan.harmon@diffio.ai)

## Remove crowd noise from interview audio: before and after

Listen for: At 0:04, the crowd in the pause before “wasn’t sure”, then the same pause restored.

Ryan Strome answers reporters at the 2011 NHL Draft, recorded on camera audio from inside the crowd. Switch between the original and the same 11.4 seconds restored with Diffio 4.7 Flash, both at −20 LUFS. Between his words the crowd is about 19 dB quieter, and his voice stays within 1 dB of its level.

[Transcript, downloads and credit](https://diffio.ai/press/#audio-media-scrum-ab)

[![Two 0 to 8 kHz spectrograms of an 11.4-second interview answer: grey crowd noise fills gaps between words on top; gaps turn white and dark voice harmonics stand out below.](https://diffio.ai/media/press/previews/remove-crowd-noise-from-interview-audio_spectrogram_1920x1080.webp)](https://diffio.ai/media/press/remove-crowd-noise-from-interview-audio_spectrogram_1920x1080.png)

**Credit: Source: "Ryan Strome, New York Islanders 2011 Draft Pick" by TermiteMound, CC BY 3.0 (creativecommons.org/licenses/by/3.0), via Wikimedia Commons. Excerpted, level-matched and speech-restored by Diffio. No endorsement implied. [License (opens in a new tab)](https://creativecommons.org/licenses/by/3.0/)**

[Download images, logos and audio (.zip)](https://diffio.ai/media/press/bundles/diffio-press-kit_images-logos-audio.zip) 31.6 MB; videos below, one file per cut

You can publish these files for free in coverage, reviews and posts without asking first. Editorial video cuts carry no Diffio logo or call to action. Keep the credit line shown with each file where there is one; Creative Commons (CC BY) footage requires it. Don't alter the original or restored audio, or present one as the other. For questions, write to Nathan Harmon at nathan.harmon@diffio.ai.

**About the measurements:** Estimated speaker similarity, quoted with several files below, is an estimate, from the noisy original and the restored file alone, of how closely the restored voice matches the speaker; 1 means identical. It and SCOREQ are machine measurements by Diffio, not a listening test, so compare the voice and words with the original before relying on a restored copy.

**On this page**

- [Videos](https://diffio.ai/press/#videos)
- [Audio](https://diffio.ai/press/#audio)
- [Before and after graphics](https://diffio.ai/press/#before-after-graphics)
- [Social cards](https://diffio.ai/press/#social-cards)
- [Product screenshots](https://diffio.ai/press/#product-screenshots)
- [Logos](https://diffio.ai/press/#logos)
- [About Diffio AI](https://diffio.ai/press/#about-diffio)
- [Fact sheet](https://diffio.ai/press/#fact-sheet)
- [Founder](https://diffio.ai/press/#founder)
- [Back to top](https://diffio.ai/press/#main-content)

## Videos

Branded

Cuts that end on the Diffio logo and “Try for free”.

Editorial

Educational cuts with no logo, URL or call to action.

### Remove crowd noise from an interview video: a media scrum

Diffio 4.7 Flash

A hockey draft pick answers reporters in the Diffio player beside a live spectrogram and captions, toggling from camera audio to Restored speech before a Diffio end card.

A draft-floor media scrum on camera audio, restored with Diffio 4.7 Flash. Hear the original, the restored speech and a spectrogram of each. Over the 15 s cut's restored stretch, the crowd between his words is about 18 dB quieter, and his voice stays within 1 dB of its level.

[This cut, MP4](https://diffio.ai/media/press/remove-crowd-noise-from-interview-video_15s.mp4) 7.5 MB

**Transcript**

I think just confidence in the second year around. I mean, I worked really hard in the summer, and I credit that to Athlete Training Center where I worked with John, obviously. And I mean, I came back with a ton of confidence. I got the ball rolling early and I just tried to keep it rolling through the whole season.

**All 6 cuts, captions and posters**

| Cut | Video | Captions | Poster |
| --- | --- | --- | --- |
| Branded, 10 s | [MP4](https://diffio.ai/media/press/remove-crowd-noise-from-interview-video_10s.mp4) 7.0 MB | [VTT](https://diffio.ai/media/press/remove-crowd-noise-from-interview-video_10s.vtt) | [PNG](https://diffio.ai/media/press/remove-crowd-noise-from-interview-video_10s.png) |
| Editorial, 10 s | [MP4](https://diffio.ai/media/press/remove-crowd-noise-from-interview-video_editorial_10s.mp4) 9.2 MB | [VTT](https://diffio.ai/media/press/remove-crowd-noise-from-interview-video_editorial_10s.vtt) | [PNG](https://diffio.ai/media/press/remove-crowd-noise-from-interview-video_editorial_10s.png) |
| Branded, 15 s | [MP4](https://diffio.ai/media/press/remove-crowd-noise-from-interview-video_15s.mp4) 7.5 MB | [VTT](https://diffio.ai/media/press/remove-crowd-noise-from-interview-video_15s.vtt) | [PNG](https://diffio.ai/media/press/remove-crowd-noise-from-interview-video_15s.png) |
| Editorial, 15 s | [MP4](https://diffio.ai/media/press/remove-crowd-noise-from-interview-video_editorial_15s.mp4) 13.5 MB | [VTT](https://diffio.ai/media/press/remove-crowd-noise-from-interview-video_editorial_15s.vtt) | [PNG](https://diffio.ai/media/press/remove-crowd-noise-from-interview-video_editorial_15s.png) |
| Branded, 30 s | [MP4](https://diffio.ai/media/press/remove-crowd-noise-from-interview-video_30s.mp4) 21.1 MB | [VTT](https://diffio.ai/media/press/remove-crowd-noise-from-interview-video_30s.vtt) | [PNG](https://diffio.ai/media/press/remove-crowd-noise-from-interview-video_30s.png) |
| Editorial, 30 s | [MP4](https://diffio.ai/media/press/remove-crowd-noise-from-interview-video_editorial_30s.mp4) 29.0 MB | [VTT](https://diffio.ai/media/press/remove-crowd-noise-from-interview-video_editorial_30s.vtt) | [PNG](https://diffio.ai/media/press/remove-crowd-noise-from-interview-video_editorial_30s.png) |

Poster: The Diffio player on a hockey draft pick at a press table, labelled Original camera audio, recorded in a media scrum, with a faint greyscale spectrogram beside the footage.

Credit: Footage: “Ryan Strome, New York Islanders 2011 Draft Pick” by TermiteMound, CC BY 3.0 (creativecommons.org/licenses/by/3.0), via Wikimedia Commons. Rotated, excerpted, level-matched and speech-restored by Diffio (Diffio 4.7 Flash). No endorsement implied. [License (opens in a new tab)](https://creativecommons.org/licenses/by/3.0/)

### Remove background noise from an outdoor interview video

Diffio 4.7 Pro

A Marine flight-line interview in the Diffio player with a live spectrogram and captions at left, toggling to Restored and 30% Background amount before a Diffio end card.

An interview recorded on a flight line, restored with Diffio 4.7 Pro. Hear the original camera audio, then the speech with 30% of the background.

Diffio 4.7 Pro, Background amount 0%: between his words the noise is about 40 dB quieter at matched loudness (clip 11.12 to 28.95 s), speech level within 1 dB. Where the label reads "Restored speech with 30% background", 30% of the flight line is mixed back in on purpose.

[This cut, MP4](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview-video_15s.mp4) 8.8 MB

**Transcript**

I also deal with our safe-for-flight program. So when it comes to the, uh, the maintenance controllers that we have and those that are actually screening aircraft and ensuring that they're prepared and ready for flight, that's a big part of my process.

**All 6 cuts, captions and posters**

| Cut | Video | Captions | Poster |
| --- | --- | --- | --- |
| Branded, 10 s | [MP4](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview-video_10s.mp4) 9.9 MB | [VTT](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview-video_10s.vtt) | [PNG](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview-video_10s.png) |
| Editorial, 10 s | [MP4](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview-video_editorial_10s.mp4) 9.9 MB | [VTT](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview-video_editorial_10s.vtt) | [PNG](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview-video_editorial_10s.png) |
| Branded, 15 s | [MP4](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview-video_15s.mp4) 8.8 MB | [VTT](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview-video_15s.vtt) | [PNG](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview-video_15s.png) |
| Editorial, 15 s | [MP4](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview-video_editorial_15s.mp4) 16.5 MB | [VTT](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview-video_editorial_15s.vtt) | [PNG](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview-video_editorial_15s.png) |
| Branded, 30 s | [MP4](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview-video_30s.mp4) 23.9 MB | [VTT](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview-video_30s.vtt) | [PNG](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview-video_30s.png) |
| Editorial, 30 s | [MP4](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview-video_editorial_30s.mp4) 32.1 MB | [VTT](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview-video_editorial_30s.vtt) | [PNG](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview-video_editorial_30s.png) |

Poster: The Diffio player on a Marine interviewed beside a parked jet on a flight line, labelled Original camera audio, recorded on a flight line, with a faint greyscale spectrogram in the margin to its left.

Credit: Footage: “Marine Attack Squadron 231 Flight Line Interviews”, U.S. Marine Corps video by Cpl. Joshua Munsen (February 13, 2025), via DVIDS. Public domain. Excerpted, markings blurred, level-matched and speech-restored by Diffio (Diffio 4.7 Pro). No endorsement by the U.S. Marine Corps implied. The appearance of U.S. Department of War (DoW) visual information does not imply or constitute DoW endorsement.

### Remove crowd noise from video: JFK at Rice, 1962

Diffio 4.7 Pro

NASA film of President John F. Kennedy speaking at a Rice University lectern in 1962, shown in the Diffio web player with a greyscale spectrogram, state labels, captions and a Background amount slider.

President Kennedy's 1962 Rice University speech, restored with Diffio 4.7 Pro. Diffio separates the stadium crowd onto a background track instead of deleting it, and in the 15-second cut the Background amount slider mixes the crowd back in at 30% and 60%. Estimated speaker similarity is 0.63, so compare his voice with the original.

[This cut, MP4](https://diffio.ai/media/press/remove-crowd-noise-from-video_jfk-rice-1962_15s.mp4) 9.4 MB

**Transcript**

Why does Rice play Texas? We choose to go to the moon. We choose to go to the moon.

**All 6 cuts, captions and posters**

| Cut | Video | Captions | Poster |
| --- | --- | --- | --- |
| Branded, 10 s | [MP4](https://diffio.ai/media/press/remove-crowd-noise-from-video_jfk-rice-1962_10s.mp4) 6.9 MB | [VTT](https://diffio.ai/media/press/remove-crowd-noise-from-video_jfk-rice-1962_10s.vtt) | [PNG](https://diffio.ai/media/press/remove-crowd-noise-from-video_jfk-rice-1962_10s.png) |
| Editorial, 10 s | [MP4](https://diffio.ai/media/press/remove-crowd-noise-from-video_jfk-rice-1962_editorial_10s.mp4) 7.6 MB | [VTT](https://diffio.ai/media/press/remove-crowd-noise-from-video_jfk-rice-1962_editorial_10s.vtt) | [PNG](https://diffio.ai/media/press/remove-crowd-noise-from-video_jfk-rice-1962_editorial_10s.png) |
| Branded, 15 s | [MP4](https://diffio.ai/media/press/remove-crowd-noise-from-video_jfk-rice-1962_15s.mp4) 9.4 MB | [VTT](https://diffio.ai/media/press/remove-crowd-noise-from-video_jfk-rice-1962_15s.vtt) | [PNG](https://diffio.ai/media/press/remove-crowd-noise-from-video_jfk-rice-1962_15s.png) |
| Editorial, 15 s | [MP4](https://diffio.ai/media/press/remove-crowd-noise-from-video_jfk-rice-1962_editorial_15s.mp4) 10.9 MB | [VTT](https://diffio.ai/media/press/remove-crowd-noise-from-video_jfk-rice-1962_editorial_15s.vtt) | [PNG](https://diffio.ai/media/press/remove-crowd-noise-from-video_jfk-rice-1962_editorial_15s.png) |
| Branded, 30 s | [MP4](https://diffio.ai/media/press/remove-crowd-noise-from-video_jfk-rice-1962_30s.mp4) 22.6 MB | [VTT](https://diffio.ai/media/press/remove-crowd-noise-from-video_jfk-rice-1962_30s.vtt) | [PNG](https://diffio.ai/media/press/remove-crowd-noise-from-video_jfk-rice-1962_30s.png) |
| Editorial, 30 s | [MP4](https://diffio.ai/media/press/remove-crowd-noise-from-video_jfk-rice-1962_editorial_30s.mp4) 23.9 MB | [VTT](https://diffio.ai/media/press/remove-crowd-noise-from-video_jfk-rice-1962_editorial_30s.vtt) | [PNG](https://diffio.ai/media/press/remove-crowd-noise-from-video_jfk-rice-1962_editorial_30s.png) |

Poster: President John F. Kennedy at the Rice University lectern in 1962, in the Diffio player, labelled Original, with a faint greyscale spectrogram and the date and place beside the film.

Credit: Film: President John F. Kennedy at Rice University, Houston, Texas, September 12, 1962. NASA, public domain, via Wikimedia Commons. Excerpted and level-matched. Speech restored with Diffio 4.7 Pro. No endorsement implied.

### Fix phone call audio for a podcast guest

Diffio 4.7 Flash

Listen for the line hiss between the words: it's there in the original and gone in the restored speech.

Full-width spectrogram card for a podcast guest on a phone line, cued "Listen for line hiss between words", over the Diffio player switching to Restored. Then a same-phrase card with the pause boxed in both halves; captions; Diffio end card.

A NOAA podcast guest on a phone line that stops at about 3.5 kHz. Hear the original, then the speech restored with Diffio 4.7 Flash. Estimated speaker similarity is 0.68 on the full 18-second passage, so check the guest's words against the original.

[This cut, MP4](https://diffio.ai/media/press/fix-phone-call-audio-for-podcast_15s.mp4) 1.7 MB

**Transcript**

Papahānaumokuākea Marine National Monument and the high levels of protection they have and the varying levels of protection

**All 6 cuts, captions and posters**

| Cut | Video | Captions | Poster |
| --- | --- | --- | --- |
| Branded, 15 s | [MP4](https://diffio.ai/media/press/fix-phone-call-audio-for-podcast_15s.mp4) 1.7 MB | [VTT](https://diffio.ai/media/press/fix-phone-call-audio-for-podcast_15s.vtt) | [PNG](https://diffio.ai/media/press/fix-phone-call-audio-for-podcast_15s.png) |
| Editorial, 15 s | [MP4](https://diffio.ai/media/press/fix-phone-call-audio-for-podcast_editorial_15s.mp4) 2.0 MB | [VTT](https://diffio.ai/media/press/fix-phone-call-audio-for-podcast_editorial_15s.vtt) | [PNG](https://diffio.ai/media/press/fix-phone-call-audio-for-podcast_editorial_15s.png) |
| Branded, 10 s | [MP4](https://diffio.ai/media/press/fix-phone-call-audio-for-podcast_10s.mp4) 1.4 MB | [VTT](https://diffio.ai/media/press/fix-phone-call-audio-for-podcast_10s.vtt) | [PNG](https://diffio.ai/media/press/fix-phone-call-audio-for-podcast_10s.png) |
| Editorial, 10 s | [MP4](https://diffio.ai/media/press/fix-phone-call-audio-for-podcast_editorial_10s.mp4) 1.4 MB | [VTT](https://diffio.ai/media/press/fix-phone-call-audio-for-podcast_editorial_10s.vtt) | [PNG](https://diffio.ai/media/press/fix-phone-call-audio-for-podcast_editorial_10s.png) |
| Branded, 25 s | [MP4](https://diffio.ai/media/press/fix-phone-call-audio-for-podcast_25s.mp4) 3.4 MB | [VTT](https://diffio.ai/media/press/fix-phone-call-audio-for-podcast_25s.vtt) | [PNG](https://diffio.ai/media/press/fix-phone-call-audio-for-podcast_25s.png) |
| Editorial, 25 s | [MP4](https://diffio.ai/media/press/fix-phone-call-audio-for-podcast_editorial_25s.mp4) 3.4 MB | [VTT](https://diffio.ai/media/press/fix-phone-call-audio-for-podcast_editorial_25s.vtt) | [PNG](https://diffio.ai/media/press/fix-phone-call-audio-for-podcast_editorial_25s.png) |

Poster: Spectrogram card for a podcast guest on a phone line, original half with the phone line cutoff marked at about 3.5 kHz, restored half on the same scale.

Credit: Audio: “Diving Deeper: Exploring Hawaii’s Coral Reefs”, NOAA Ocean Service podcast, December 2015, public domain. Excerpted and level-matched. Speech restored with Diffio 4.7 Flash. No endorsement by NOAA or the speakers is implied.

### Clean up audio from a video call recording: a USGS talk

Diffio 4.7 Flash

A geologist speaks to his webcam in a lecture given over a video call, with captions beside him, Original and Restored speech labels and a greyscale spectrogram below, before a Diffio end card.

A USGS talk given over a video call, with a noise bed under every word. Hear the original, then the speech restored by Diffio 4.7 Flash. On the 15 s cut's restored window, model-predicted SCOREQ rises from about 2.8 to about 4.7, and estimated speaker similarity on the full passage is 0.79.

[This cut, MP4](https://diffio.ai/media/press/clean-up-audio-from-video-call-recording_15s.mp4) 3.6 MB

**Transcript**

Did we really know about these faults? Did we really know about these faults? Their work after the Ridgecrest earthquake evaluated how much of the 2019 ruptures were or could have been known beforehand.

**All 6 cuts, captions and posters**

| Cut | Video | Captions | Poster |
| --- | --- | --- | --- |
| Branded, 10 s | [MP4](https://diffio.ai/media/press/clean-up-audio-from-video-call-recording_10s.mp4) 2.4 MB | [VTT](https://diffio.ai/media/press/clean-up-audio-from-video-call-recording_10s.vtt) | [PNG](https://diffio.ai/media/press/clean-up-audio-from-video-call-recording_10s.png) |
| Editorial, 10 s | [MP4](https://diffio.ai/media/press/clean-up-audio-from-video-call-recording-editorial_10s.mp4) 2.4 MB | [VTT](https://diffio.ai/media/press/clean-up-audio-from-video-call-recording-editorial_10s.vtt) | [PNG](https://diffio.ai/media/press/clean-up-audio-from-video-call-recording-editorial_10s.png) |
| Branded, 15 s | [MP4](https://diffio.ai/media/press/clean-up-audio-from-video-call-recording_15s.mp4) 3.6 MB | [VTT](https://diffio.ai/media/press/clean-up-audio-from-video-call-recording_15s.vtt) | [PNG](https://diffio.ai/media/press/clean-up-audio-from-video-call-recording_15s.png) |
| Editorial, 15 s | [MP4](https://diffio.ai/media/press/clean-up-audio-from-video-call-recording-editorial_15s.mp4) 3.6 MB | [VTT](https://diffio.ai/media/press/clean-up-audio-from-video-call-recording-editorial_15s.vtt) | [PNG](https://diffio.ai/media/press/clean-up-audio-from-video-call-recording-editorial_15s.png) |
| Branded, 27 s | [MP4](https://diffio.ai/media/press/clean-up-audio-from-video-call-recording_30s.mp4) 6.1 MB | [VTT](https://diffio.ai/media/press/clean-up-audio-from-video-call-recording_30s.vtt) | [PNG](https://diffio.ai/media/press/clean-up-audio-from-video-call-recording_30s.png) |
| Editorial, 30 s | [MP4](https://diffio.ai/media/press/clean-up-audio-from-video-call-recording-editorial_30s.mp4) 7.1 MB | [VTT](https://diffio.ai/media/press/clean-up-audio-from-video-call-recording-editorial_30s.vtt) | [PNG](https://diffio.ai/media/press/clean-up-audio-from-video-call-recording-editorial_30s.png) |

Poster: A geologist mid-sentence on a webcam, labelled Restored speech, with the title “Talk given over a video call” and a caption beside him and a spectrogram strip below.

Credit: Audio and webcam: “Where Earthquakes Hide in the Desert”, USGS Public Lecture Series, May 27, 2021, public domain. A talk given over a video call. Excerpted and speech-restored by Diffio (Diffio 4.7 Flash). No endorsement by the U.S. Geological Survey or the speaker is implied.

### Remove echo from interview video: a school cafeteria, before and after

Diffio 4.7 Pro

VOA News footage of a Red Cross volunteer interviewed in a school cafeteria used as a hurricane shelter, with Original and Restored speech labels, captions and a greyscale spectrogram before a Diffio end card.

An interview recorded in an echoey school cafeteria. Hear the original, then the speech restored by Diffio 4.7 Pro at the same loudness. Over the 10.1-second answer the audio sample and the 15 s cut play, the level between his words is about 17 dB lower (room noise and echo tail together), and his voice stays within 1 dB of its level.

[This cut, MP4](https://diffio.ai/media/press/remove-echo-from-video-cafeteria-interview_15s.mp4) 10.6 MB

**Transcript**

And I'm working here at the Smith Middle School in Chapel Hill, North Carolina… Smith Middle School in Chapel Hill, North Carolina, where we have a setup in collaboration with the county, Orange County of North Carolina.

**All 6 cuts, captions and posters**

| Cut | Video | Captions | Poster |
| --- | --- | --- | --- |
| Branded, 10 s | [MP4](https://diffio.ai/media/press/remove-echo-from-video-cafeteria-interview_10s.mp4) 7.5 MB | [VTT](https://diffio.ai/media/press/remove-echo-from-video-cafeteria-interview_10s.vtt) | [PNG](https://diffio.ai/media/press/remove-echo-from-video-cafeteria-interview_10s.png) |
| Editorial, 10 s | [MP4](https://diffio.ai/media/press/remove-echo-from-video-cafeteria-interview-editorial_10s.mp4) 7.5 MB | [VTT](https://diffio.ai/media/press/remove-echo-from-video-cafeteria-interview-editorial_10s.vtt) | [PNG](https://diffio.ai/media/press/remove-echo-from-video-cafeteria-interview-editorial_10s.png) |
| Branded, 15 s | [MP4](https://diffio.ai/media/press/remove-echo-from-video-cafeteria-interview_15s.mp4) 10.6 MB | [VTT](https://diffio.ai/media/press/remove-echo-from-video-cafeteria-interview_15s.vtt) | [PNG](https://diffio.ai/media/press/remove-echo-from-video-cafeteria-interview_15s.png) |
| Editorial, 15 s | [MP4](https://diffio.ai/media/press/remove-echo-from-video-cafeteria-interview-editorial_15s.mp4) 10.6 MB | [VTT](https://diffio.ai/media/press/remove-echo-from-video-cafeteria-interview-editorial_15s.vtt) | [PNG](https://diffio.ai/media/press/remove-echo-from-video-cafeteria-interview-editorial_15s.png) |
| Branded, 30 s | [MP4](https://diffio.ai/media/press/remove-echo-from-video-cafeteria-interview_30s.mp4) 19.9 MB | [VTT](https://diffio.ai/media/press/remove-echo-from-video-cafeteria-interview_30s.vtt) | [PNG](https://diffio.ai/media/press/remove-echo-from-video-cafeteria-interview_30s.png) |
| Editorial, 30 s | [MP4](https://diffio.ai/media/press/remove-echo-from-video-cafeteria-interview-editorial_30s.mp4) 22.3 MB | [VTT](https://diffio.ai/media/press/remove-echo-from-video-cafeteria-interview-editorial_30s.vtt) | [PNG](https://diffio.ai/media/press/remove-echo-from-video-cafeteria-interview-editorial_30s.png) |

Poster: A Red Cross volunteer at a cafeteria table, labelled Restored speech and “Interview recorded in a school cafeteria”, with a caption and a spectrogram strip below.

Credit: Video: “Red Cross Shelter Serves Florence Evacuees”, VOA News, September 14, 2018, public domain. Excerpted and speech-restored by Diffio (Diffio 4.7 Pro). No endorsement by VOA or the speaker is implied.

### Restore old sermon recordings: hear a 1969 sermon restored

Diffio 4.7 Pro

Archive band for a 1969 sermon with a small open-reel tape photo labelled "Illustration", its pill turning from "Muffled archive copy" to "Archive copy", above a full-width 0 to 4 kHz spectrogram and captions; Diffio end card.

A 1969 sermon by T. Ernest Wilson, as the archive copy sounds and restored with Diffio 4.7 Pro, level-matched. Estimated speaker similarity on the full passage is 0.72 for the 10 s and 15 s cuts and 0.70 for the 30 s, so compare the preacher's voice with the archive copy.

The video and the audio sample come from different passages of the same 1969 sermon: the video's passage reaches about 4 kHz and plays Diffio 4.7 Pro; the sample's reaches about 3 kHz and plays Diffio 4.7 Flash.

[This cut, MP4](https://diffio.ai/media/press/restore-old-sermon-recordings_30s.mp4) 2.8 MB

**Transcript**

And on the Sunday afternoon, we carried her down to the African forest, where four other missionaries are buried, and we laid her there until the daybreak, and the shadows flee away. You think I can ever forget this verse? Casting all your care upon him, for he cares for you. My beloved friend, I don't know what your care is this morning. Your anxious care. It could be anything at all.

**All 6 cuts, captions and posters**

| Cut | Video | Captions | Poster |
| --- | --- | --- | --- |
| Branded, 10 s | [MP4](https://diffio.ai/media/press/restore-old-sermon-recordings_10s.mp4) 1.2 MB | [VTT](https://diffio.ai/media/press/restore-old-sermon-recordings_10s.vtt) | [PNG](https://diffio.ai/media/press/restore-old-sermon-recordings_10s.png) |
| Editorial, 10 s | [MP4](https://diffio.ai/media/press/restore-old-sermon-recordings_editorial_10s.mp4) 1.2 MB | [VTT](https://diffio.ai/media/press/restore-old-sermon-recordings_editorial_10s.vtt) | [PNG](https://diffio.ai/media/press/restore-old-sermon-recordings_editorial_10s.png) |
| Branded, 15 s | [MP4](https://diffio.ai/media/press/restore-old-sermon-recordings_15s.mp4) 1.6 MB | [VTT](https://diffio.ai/media/press/restore-old-sermon-recordings_15s.vtt) | [PNG](https://diffio.ai/media/press/restore-old-sermon-recordings_15s.png) |
| Editorial, 15 s | [MP4](https://diffio.ai/media/press/restore-old-sermon-recordings_editorial_15s.mp4) 1.6 MB | [VTT](https://diffio.ai/media/press/restore-old-sermon-recordings_editorial_15s.vtt) | [PNG](https://diffio.ai/media/press/restore-old-sermon-recordings_editorial_15s.png) |
| Branded, 30 s | [MP4](https://diffio.ai/media/press/restore-old-sermon-recordings_30s.mp4) 2.8 MB | [VTT](https://diffio.ai/media/press/restore-old-sermon-recordings_30s.vtt) | [PNG](https://diffio.ai/media/press/restore-old-sermon-recordings_30s.png) |
| Editorial, 30 s | [MP4](https://diffio.ai/media/press/restore-old-sermon-recordings_editorial_30s.mp4) 2.7 MB | [VTT](https://diffio.ai/media/press/restore-old-sermon-recordings_editorial_30s.vtt) | [PNG](https://diffio.ai/media/press/restore-old-sermon-recordings_editorial_30s.png) |

Poster: Archive band for a February 13, 1969 sermon by T. Ernest Wilson with an open-reel tape photo and the pill "Archive copy, 1969", under "Restored speech", above a full-width spectrogram and caption.

Credit: Sermon: T. Ernest Wilson, "The God Of All Grace" (1 Peter 5:1-11), February 13, 1969. Recording: Voices For Christ (vfc.diffio.ai/sermons/52421). Restored with Diffio 4.7 Pro. Illustration: open-reel tape recorder, photo by Toki-ho, Wikimedia Commons, public domain.

### Product tour video, sermon

Diffio 4.7 Pro

Compare, then read the transcript.

A tour of the Diffio app on a 1969 sermon: it opens close on the player's toggle, which switches to Restored, then pulls back to an archive card and the player, with the transcript following the words. Captions; end card.

A 1969 sermon restored in the Diffio app: switch to the restored speech, then read the transcript. One player switches between the original and restored speech, and the transcript follows the words as they play; estimated speaker similarity on the full passage is 0.72, so compare the preacher's voice with the archive copy.

[This cut, MP4](https://diffio.ai/media/press/how-to-restore-a-sermon-recording_tour_15s.mp4) 1.4 MB

**Transcript**

And on the Sunday afternoon, we carried her down to the African forest, where four other missionaries are buried, and we laid her there until the daybreak, and the shadows flee away. You think I can ever forget this verse?

**The cut, its captions and poster**

| Cut | Video | Captions | Poster |
| --- | --- | --- | --- |
| Branded, 15 s | [MP4](https://diffio.ai/media/press/how-to-restore-a-sermon-recording_tour_15s.mp4) 1.4 MB | [VTT](https://diffio.ai/media/press/how-to-restore-a-sermon-recording_tour_15s.vtt) | [PNG](https://diffio.ai/media/press/how-to-restore-a-sermon-recording_tour_15s.png) |

Poster: Archive card for a February 13, 1969 sermon by T. Ernest Wilson, marked "Archive copy, 1969", beside the Diffio player switched to RESTORED, under "The God of All Grace" and the step "Switch between original and restored".

Credit: Sermon: T. Ernest Wilson, "The God Of All Grace" (1 Peter 5:1-11), February 13, 1969. Recording: Voices For Christ (vfc.diffio.ai/sermons/52421). Restored with Diffio 4.7 Pro. App footage: Diffio.

### 1940 oral history, original and restored

Diffio 4.7 Flash

Listen for the surface hiss between words.

A 1940 camp-recording photo beside "The Johnsons' wedding day" card and spectrogram, the Diffio app player flipping Original to Restored, side-by-side Original and Restored spectrograms with captions and a Diffio end card.

“And we had a big table, a long table, out on our back porch”: from an interview about the Johnsons' wedding day, recorded on disc by Charles Todd and Robert Sonkin at the Yuba City FSA Camp on August 18, 1940, now in the Library of Congress's Todd and Sonkin migrant workers collection (AFC 1985/001). Oral-history collections with field discs, open-reel tapes or cassettes use Diffio the same way: flip in the app from the original transfer to speech restored with Diffio 4.7 Flash, compare the closing phrase both ways and publish a listening copy while the master stays untouched.

[This cut, MP4](https://diffio.ai/media/press/oral-history-audio-restoration-1940_15s.mp4) 5.5 MB

**Transcript**

And we had a big table, a long table, out on our back porch, and had it just loaded with everything you could think of. …and had it just loaded with everything you could think of.

**All 2 cuts, captions and posters**

| Cut | Video | Captions | Poster |
| --- | --- | --- | --- |
| Branded, 15 s | [MP4](https://diffio.ai/media/press/oral-history-audio-restoration-1940_15s.mp4) 5.5 MB | [VTT](https://diffio.ai/media/press/oral-history-audio-restoration-1940_15s.vtt) | [PNG](https://diffio.ai/media/press/oral-history-audio-restoration-1940_15s.png) |
| Editorial, 15 s | [MP4](https://diffio.ai/media/press/oral-history-audio-restoration-1940_editorial_15s.mp4) 7.4 MB | [VTT](https://diffio.ai/media/press/oral-history-audio-restoration-1940_editorial_15s.vtt) | [PNG](https://diffio.ai/media/press/oral-history-audio-restoration-1940_editorial_15s.png) |

Poster: The Diffio app's player in Restored mode at 0:53 of the 1940 oral history, Background amount at 0%, with "In the Diffio app" and "Restored speech" labels and the caption "out on our back porch".

Credit: Recording: "Interview about the Johnson's wedding day", Yuba City FSA Camp, August 18, 1940. Photo: Charles L. Todd and Robert Sonkin recording at Arvin FSA Camp, 1940, by Robert Hemmig (AFC 1985/001: p008; not the Johnson interview). Charles L. Todd and Robert Sonkin migrant workers collection (AFC 1985/001), American Folklife Center, Library of Congress. No known restrictions. Restored speech: Diffio 4.7 Flash. No endorsement implied.

### Restore an old audio recording: Roosevelt 1912 and Edison 1919

Diffio 4.7 Flash

Cards and spectrograms for Roosevelt's 1912 cylinder and Edison's 1919 disc, the Diffio player on Restored, side-by-side spectrograms with subhead "For oral-history archives, museums and libraries with old transfers" and a Diffio end card.

Theodore Roosevelt's August 1912 Edison cylinder and Thomas Edison's 1919 disc sit under heavy surface hiss in their Library of Congress transfers. For oral-history archives, museums and libraries with old transfers, Diffio 4.7 Flash makes a separate listening copy while keeping the archival transfer as the record. On recordings this worn, estimated speaker similarity is 0.62 (Roosevelt) and 0.58 (Edison), so check the voice and words against the transfers. For a less damaged disc recording, compare [the 1940 interview](https://diffio.ai/press/#video-oral-history-tour) from a farm workers' camp. Recordings and photos: Library of Congress.

Estimated speaker similarity of the restored speech: 0.62 (Roosevelt) and 0.58 (Edison). Check the voice and words against the Library of Congress transfers before citing the restored copy. [What this measures](https://diffio.ai/press/#measurements)

[This cut, MP4](https://diffio.ai/media/press/restore-old-audio-recording_roosevelt-edison_30s.mp4) 7.3 MB

**Transcript**

I believe in the right of the people to rule. I believe that the majority of the plain people of the United States will, day in and day out, make fewer mistakes in governing themselves. I hope that when we do reverence to the memory of our brave boys who fell in France. …day in and day out, make fewer mistakes in governing themselves. …day in and day out, make fewer mistakes in governing themselves.

**All 6 cuts, captions and posters**

| Cut | Video | Captions | Poster |
| --- | --- | --- | --- |
| Branded, 10 s | [MP4](https://diffio.ai/media/press/restore-old-audio-recording_roosevelt-1912_10s.mp4) 2.9 MB | [VTT](https://diffio.ai/media/press/restore-old-audio-recording_roosevelt-1912_10s.vtt) | [PNG](https://diffio.ai/media/press/restore-old-audio-recording_roosevelt-1912_10s.png) |
| Editorial, 10 s | [MP4](https://diffio.ai/media/press/restore-old-audio-recording_roosevelt-1912_editorial_10s.mp4) 2.9 MB | [VTT](https://diffio.ai/media/press/restore-old-audio-recording_roosevelt-1912_editorial_10s.vtt) | [PNG](https://diffio.ai/media/press/restore-old-audio-recording_roosevelt-1912_editorial_10s.png) |
| Branded, 15 s | [MP4](https://diffio.ai/media/press/restore-old-audio-recording_roosevelt-1912_15s.mp4) 3.1 MB | [VTT](https://diffio.ai/media/press/restore-old-audio-recording_roosevelt-1912_15s.vtt) | [PNG](https://diffio.ai/media/press/restore-old-audio-recording_roosevelt-1912_15s.png) |
| Editorial, 15 s | [MP4](https://diffio.ai/media/press/restore-old-audio-recording_roosevelt-1912_editorial_15s.mp4) 3.5 MB | [VTT](https://diffio.ai/media/press/restore-old-audio-recording_roosevelt-1912_editorial_15s.vtt) | [PNG](https://diffio.ai/media/press/restore-old-audio-recording_roosevelt-1912_editorial_15s.png) |
| Branded, 30 s | [MP4](https://diffio.ai/media/press/restore-old-audio-recording_roosevelt-edison_30s.mp4) 7.3 MB | [VTT](https://diffio.ai/media/press/restore-old-audio-recording_roosevelt-edison_30s.vtt) | [PNG](https://diffio.ai/media/press/restore-old-audio-recording_roosevelt-edison_30s.png) |
| Editorial, 30 s | [MP4](https://diffio.ai/media/press/restore-old-audio-recording_roosevelt-edison_editorial_30s.mp4) 7.9 MB | [VTT](https://diffio.ai/media/press/restore-old-audio-recording_roosevelt-edison_editorial_30s.vtt) | [PNG](https://diffio.ai/media/press/restore-old-audio-recording_roosevelt-edison_editorial_30s.png) |

Poster: Black-and-white photo of Theodore Roosevelt beside "Heard twice: original, then restored", side-by-side Original and Restored speech spectrograms and subhead "For oral-history archives, museums and libraries with old transfers".

Credit: Recordings: Theodore Roosevelt, "The right of the people to rule", Edison cylinder, 1912 (Library of Congress, item 99391599) and Thomas A. Edison, "Let us not forget", Edison Diamond Disc 50509, 1919 (Library of Congress, item 00694069). Photos: Library of Congress (LC-DIG-ppmsca-36659, LC-USZ62-91115). Public domain in the US. Restored speech: Diffio 4.7 Flash. No endorsement implied.

### Restore an old radio recording: Amelia Earhart, 1935

Diffio 4.7 Pro

A 1937 black-and-white photo of Amelia Earhart beside the label "Damaged radio transfer · listening copy", "Old radio broadcast, 1935", captions, a spectrogram labelled Original, Restored speech and Original again, the note "Estimated speaker similarity 0.58", then a Diffio end card.

**Note:** A person has not yet listened to this cut, and Diffio's founder has not yet signed it off; listen before you use it.

For radio and broadcast archives holding damaged air-check transfers: a severely damaged 1935 radio transfer, restored as a listening copy. Compare her voice and words with the original before relying on the restored copy. In the broadcast, "A Woman's Place in Science", Amelia Earhart's voice comes through hiss and heavy distortion. Diffio 4.7 Pro restores her opening words, then the sentence finishes as recorded, so you can judge both side by side while the source recording stays the record. Estimated speaker similarity on the full passage is 0.58. Most archival recordings aren't famous broadcasts: for everyday collection transfers with less broadcast damage, compare the [1940 camp oral history, original and restored](https://diffio.ai/press/#video-oral-history-tour) (speaker similarity 0.75).

Estimated speaker similarity 0.58 on the full passage. On a broadcast this damaged, compare her voice with the original before relying on the restored copy. [What this measures](https://diffio.ai/press/#measurements)

[This cut, MP4](https://diffio.ai/media/press/restore-old-radio-recording_earhart-1935_10s.mp4) 2.3 MB

**Transcript**

This modern world… This modern world of science and invention is of particular interest to women.

**All 2 cuts, captions and posters**

| Cut | Video | Captions | Poster |
| --- | --- | --- | --- |
| Branded, 10 s | [MP4](https://diffio.ai/media/press/restore-old-radio-recording_earhart-1935_10s.mp4) 2.3 MB | [VTT](https://diffio.ai/media/press/restore-old-radio-recording_earhart-1935_10s.vtt) | [PNG](https://diffio.ai/media/press/restore-old-radio-recording_earhart-1935_10s.png) |
| Editorial, 10 s | [MP4](https://diffio.ai/media/press/restore-old-radio-recording_earhart-1935_editorial_10s.mp4) 2.3 MB | [VTT](https://diffio.ai/media/press/restore-old-radio-recording_earhart-1935_editorial_10s.vtt) | [PNG](https://diffio.ai/media/press/restore-old-radio-recording_earhart-1935_editorial_10s.png) |

Poster: Black-and-white 1937 photo of Amelia Earhart smiling beside "Restored speech", "Amelia Earhart", "Old radio broadcast, 1935", a split spectrogram (grey hiss under Original, dark speech bands under Restored speech) and the note "Estimated speaker similarity 0.58".

Credit: Audio: Amelia Earhart, "A Woman's Place in Science", radio broadcast, 1935. Photo: Underwood & Underwood, 1937, Smithsonian National Air and Space Museum, public domain. Restored speech: Diffio 4.7 Pro. No endorsement implied.

### Speech restoration API in Python, run on a noisy file

Diffio 4.7 Flash

Terminal running a nine-line Python script against the Diffio API, followed by Original and Restored hatchery spectrograms. Status bar: "Recorded with diffio 1.0.0 Python SDK (API v2)".

pip install diffio, nine lines of Python and a recording full of machine noise, restored by Diffio 4.7 Flash on the live API. The script calls the model `diffio-flash` (the docs' alias for `diffio-4.7-flash`), which the API bills at $5 per hour of source audio (about $0.001389 per second, billed per second with a 60-second minimum per job), and a new API account starts with a one-time $5 credit, no card needed. This run took 19.2 seconds for the 23.6-second recording, including upload and download (one run, not a speed benchmark). Recorded with diffio 1.0.0 Python SDK (API v2). Hear the before and after.

**Version note:** Recorded with diffio 1.0.0 Python SDK (API v2). [Current quickstart](https://diffio.ai/docs/developers/quick-start/)

[This cut, MP4](https://diffio.ai/media/press/speech-restoration-api-python_27s.mp4) 4.8 MB

**Code in the video**

**Version note:** Recorded with diffio 1.0.0 Python SDK (API v2).

Terminal

```
pip install diffio
export DIFFIO_API_KEY="diffio_live_..."   # your key; the video only checks that it's set and never shows it
python restore.py
```

restore.py (diffio 1.0.0, API v2), as it ran in the video

```
from diffio import DiffioClient

client = DiffioClient()  # DIFFIO_API_KEY
result = client.restore_audio(
    "hatchery.wav",
    model="diffio-flash",
    output_path="restored.mp3",
    on_progress=lambda g: print(g.stage),
)
```

[Explore SDK docs](https://diffio.ai/docs/developers/quick-start/)

**Transcript**

As the filter panels clog up, in the center of the filter panel, the water rises up as the filter clogs. As the filter panels clog up, in the center of the filter panel… As the filter panels clog up, in the center of the filter panel, the water rises up as the filter clogs. Eventually it kicks on a float switch.

**All 3 cuts, captions and posters**

| Cut | Video | Captions | Poster |
| --- | --- | --- | --- |
| Branded, 27 s, 16:9 | [MP4](https://diffio.ai/media/press/speech-restoration-api-python_27s.mp4) 4.8 MB | [VTT](https://diffio.ai/media/press/speech-restoration-api-python_27s.vtt) | [PNG](https://diffio.ai/media/press/speech-restoration-api-python_27s_poster.png) |
| Branded, 27 s, square | [MP4](https://diffio.ai/media/press/speech-restoration-api-python_27s_square.mp4) 3.4 MB | [VTT](https://diffio.ai/media/press/speech-restoration-api-python_27s_square.vtt) | [PNG](https://diffio.ai/media/press/speech-restoration-api-python_27s_square_poster.png) |
| Editorial, 27 s | [MP4](https://diffio.ai/media/press/speech-restoration-api-python-editorial_27s.mp4) 4.8 MB | [VTT](https://diffio.ai/media/press/speech-restoration-api-python-editorial_27s.vtt) | [PNG](https://diffio.ai/media/press/speech-restoration-api-python-editorial_27s_poster.png) |

Poster: Stacked spectrograms of Original (hatchery.wav) machine noise and Restored speech (restored.mp3) harmonics, captioned "in the center of the filter panel", above "Recorded with diffio 1.0.0 Python SDK (API v2)".

Credit: Recording: "Filtration Video at Leadville NFH", Stephanie Raine, U.S. Fish and Wildlife Service, public domain (US federal work, via Wikimedia Commons). Excerpted, level-matched and speech-restored by Diffio.

## Before and after audio

### Remove crowd noise from interview audio: before and after

Listen for: At 0:04, the crowd in the pause before “wasn’t sure”, then the same pause restored.

Original![Spectrogram of the scrum answer as recorded, 0 to 8 kHz, crowd noise throughout.](https://diffio.ai/media/press/previews/media-scrum_spectrogram-original_1600x360.webp)

Restored speech![Spectrogram of the same answer restored with Diffio 4.7 Flash, 0 to 8 kHz.](https://diffio.ai/media/press/previews/media-scrum_spectrogram-restored-diffio-4.7-flash_1600x360.webp)

**Spectrograms, 0 to 8 kHz, 11.4 s, on one dB scale; the line at 0:04 marks the “Listen for” moment.**

[Restored speech, Diffio 4.7 Flash, 11.4 s, WAV](https://diffio.ai/media/press/media-scrum-interview_restored-diffio-4.7-flash_11s.wav) 2.1 MB

Ryan Strome answers reporters at the 2011 NHL Draft, recorded on camera audio from inside the crowd. Switch between the original and the same 11.4 seconds restored with Diffio 4.7 Flash, both at −20 LUFS. Between his words the crowd is about 19 dB quieter, and his voice stays within 1 dB of its level.

An 11.4-second answer, restored with Diffio 4.7 Flash and level-matched to the original. Between his words the crowd is about 19 dB quieter and his voice stays within 1 dB of its level. Machine measurements, not a listening test. In the restored half, “said” at 0:19 may sound like “sit”; check it against the original at 0:06.

- [Before and after, MP3](https://diffio.ai/media/press/remove-crowd-noise-from-interview-audio_before-after.mp3) 935 KB Restored speech from 0:12.4 (Diffio 4.7 Flash) · Original 0:00 to 0:11
- [Original, 11.4 s, WAV](https://diffio.ai/media/press/media-scrum-interview_original_11s.wav) 2.1 MB
- [Restored speech, Diffio 4.7 Pro, 11.4 s, WAV](https://diffio.ai/media/press/media-scrum-interview_restored-diffio-4.7-pro_11s.wav) 2.1 MB

**Transcript**

Uh, not until they announced after Niagara IceDogs. I mean, Dougie Hamilton's still on the board, so I mean, uh, I wasn't sure, and uh, I mean, I just said the whole time, but uh, I mean, I, I was just so thrilled that I really can't remember much of it.

Credit: Source: "Ryan Strome, New York Islanders 2011 Draft Pick" by TermiteMound, CC BY 3.0 (creativecommons.org/licenses/by/3.0), via Wikimedia Commons. Excerpted, level-matched and speech-restored by Diffio. No endorsement implied. [License (opens in a new tab)](https://creativecommons.org/licenses/by/3.0/)

### Remove echo from interview audio: before and after

Listen for: At 0:04, the cafeteria’s echo tail after “North Carolina”, then the same pause restored.

Original![Spectrogram of the echo-room answer as recorded in a school cafeteria, 0 to 8 kHz, grey room noise and echo tail between words.](https://diffio.ai/media/press/previews/echo-room_spectrogram-original_1600x360.webp)

Restored speech![Spectrogram of the same answer restored with Diffio 4.7 Pro, 0 to 8 kHz.](https://diffio.ai/media/press/previews/echo-room_spectrogram-restored-diffio-4.7-pro_1600x360.webp)

**Spectrograms, 0 to 8 kHz, 10.1 s, on one dB scale; the line at 0:04 marks the “Listen for” moment.**

[Restored speech, Diffio 4.7 Pro, 10.1 s, WAV](https://diffio.ai/media/press/echo-room-interview_restored-diffio-4.7-pro_10s.wav) 1.8 MB

A Red Cross volunteer interviewed by VOA News in a middle-school cafeteria used as a hurricane shelter, 2018. Switch between the original and the same 10.1 seconds restored with Diffio 4.7 Pro, both at −20 LUFS. Between his words the level is about 17 dB lower (room noise and echo tail together), and his voice stays within 1 dB of its level.

A 10.1-second answer from a VOA interview in a school cafeteria, restored with Diffio 4.7 Pro and level-matched. Between his words the level is about 17 dB lower (room noise and echo tail together), and his voice stays within 1 dB of its level. Machine measurements, not a listening test.

- [Before and after, MP3](https://diffio.ai/media/press/remove-echo-from-interview-audio_before-after.mp3) 834 KB Restored speech from 0:11.1 (Diffio 4.7 Pro) · Original 0:00 to 0:10
- [Original, 10.1 s, WAV](https://diffio.ai/media/press/echo-room-interview_original_10s.wav) 1.8 MB

**Transcript**

And I'm working here at the Smith Middle School in Chapel Hill, North Carolina, where we have a setup in collaboration with the county, Orange County of North Carolina.

Credit: Video: VOA News, "Red Cross Shelter Serves Florence Evacuees" (September 14, 2018), public domain. Excerpted, level-matched and speech-restored by Diffio. [License (opens in a new tab)](https://creativecommons.org/publicdomain/mark/1.0/)

### Remove noise from an outdoor interview: before and after

Listen for: At 0:01, the jet and ground noise under “maintenance controllers”, then that passage restored.

Original![Spectrogram of the flight line answer as recorded, 0 to 8 kHz, jet and ground noise throughout](https://diffio.ai/media/press/previews/flight-line_spectrogram-original_1600x360.webp)

Restored speech![Spectrogram of the same passage restored with Diffio 4.7 Pro, 0 to 8 kHz](https://diffio.ai/media/press/previews/flight-line_spectrogram-restored-diffio-4.7-pro_1600x360.webp)

**Spectrograms, 0 to 8 kHz, 10.5 s, on one dB scale; the line at 0:01 marks the “Listen for” moment.**

[Restored speech, Diffio 4.7 Pro, 10.5 s, WAV](https://diffio.ai/media/press/flight-line-interview_restored-diffio-4.7-pro_11s.wav) 1.9 MB

A U.S. Marine Corps maintenance chief interviewed on a flight line in 2025. Switch between the original camera audio and the same 10.5 seconds restored with Diffio 4.7 Pro, both at −20 LUFS.

A 10.5-second sentence recorded on a flight line, restored with Diffio 4.7 Pro in the Diffio web app and level-matched. SCOREQ 2.86 → 5.06 (model-predicted; the scale is nominally 1 to 5 and the model's predictions can run slightly past 5) and estimated speaker similarity 0.76; his voice stays within 1 dB of its level. Machine measurements, not a listening test.

- [Before and after, MP3](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview_before-after.mp3) 866 KB Restored speech from 0:11.5 (Diffio 4.7 Pro) · Original 0:00 to 0:10.5
- [Original, 10.5 s, WAV](https://diffio.ai/media/press/flight-line-interview_original_11s.wav) 1.9 MB

**Transcript**

So when it comes to the, uh, the maintenance controllers that we have and those that are actually screening aircraft and ensuring that they're prepared and ready for flight, that's a big part of my process.

Credit: Footage: "Marine Attack Squadron 231 Flight Line Interviews", U.S. Marine Corps video by Cpl. Joshua Munsen (February 13, 2025), via DVIDS. Public domain. Excerpted, level-matched and speech-restored by Diffio. The appearance of U.S. Department of War (DoW) visual information does not imply or constitute DoW endorsement. [License (opens in a new tab)](https://creativecommons.org/publicdomain/mark/1.0/)

### Restore an old sermon recording: before and after

Listen for: At 0:04, “2,000 Africans”, where the archived copy’s voice breaks up (outlined on the spectrogram), then the same passage restored.

Original![Spectrogram of the archived sermon passage, 0 to 3 kHz.](https://diffio.ai/media/press/previews/old-sermon_spectrogram-original_1600x360.webp)

Restored speech![Spectrogram of the same passage restored with Diffio 4.7 Flash, 0 to 3 kHz.](https://diffio.ai/media/press/previews/old-sermon_spectrogram-restored-diffio-4.7-flash_1600x360.webp)

**Spectrograms, 0 to 3 kHz, 13.4 s, on one dB scale; the line at 0:04 marks the “Listen for” moment.**

[Restored speech, Diffio 4.7 Flash, 13.4 s, WAV](https://diffio.ai/media/press/old-sermon-recording_restored-diffio-4.7-flash_13s.wav) 2.5 MB

T. Ernest Wilson, "The God Of All Grace", February 13, 1969, from the Voices For Christ archive. Switch between the archived copy and the same 13.4 seconds restored with Diffio 4.7 Flash, both at −20 LUFS. Machine transcripts of the two agree on all 37 words, but Diffio's restore is guided by a transcript, so check the words against the archived copy.

A 13.4-second passage from an archived copy that holds speech up to about 3 kHz, restored with Diffio 4.7 Flash and level-matched. Machine transcripts of the two agree on all 37 words, and estimated speaker similarity is 0.74, so check the voice and words against the archived copy. Machine measurements, not a listening test; Diffio's restore is guided by a transcript, so the transcript match doesn't prove no word changed.

The video and the audio sample come from different passages of the same 1969 sermon: the video's passage reaches about 4 kHz and plays Diffio 4.7 Pro; the sample's reaches about 3 kHz and plays Diffio 4.7 Flash.

- [Before and after, MP3](https://diffio.ai/media/press/restore-old-sermon-recording_before-after.mp3) 1.1 MB Restored speech from 0:14.4 (Diffio 4.7 Flash) · Archived copy 0:00 to 0:13
- [Archived copy, 13.4 s, WAV](https://diffio.ai/media/press/old-sermon-recording_original_13s.wav) 2.5 MB
- [Restored speech, Diffio 4.7 Pro, 13.4 s, WAV](https://diffio.ai/media/press/old-sermon-recording_restored-diffio-4.7-pro_13s.wav) 2.5 MB

**Transcript**

And that afternoon, we had a meeting, we had a funeral service with about 2,000 Africans, and I think 14 white men were there, Portuguese officials and traders, and we had a very lovely funeral service.

Credit: Sermon: T. Ernest Wilson, "The God Of All Grace", February 13, 1969, Voices For Christ (vfc.diffio.ai/sermons/52421). Excerpted, level-matched and speech-restored by Diffio.

### Oral history audio restoration: before and after

Listen for: At 0:01, the hiss of the 1940 disc under “back porch”, then that passage restored.

Original![Spectrogram of the 1940 interview as recorded, 0 to 4 kHz, disc hiss throughout](https://diffio.ai/media/press/previews/oral-history-1940_spectrogram-original_1600x360.webp)

Restored speech![Spectrogram of the same passage restored with Diffio 4.7 Flash, 0 to 4 kHz](https://diffio.ai/media/press/previews/oral-history-1940_spectrogram-restored-diffio-4.7-flash_1600x360.webp)

**Spectrograms, 0 to 4 kHz, 5.35 s, on one dB scale; the line at 0:01 marks the “Listen for” moment.**

[Restored speech, Diffio 4.7 Flash, 5.35 s, WAV](https://diffio.ai/media/press/oral-history-1940_restored-diffio-4.7-flash_5s.wav) 1003 KB

A wedding-day memory recorded at the Yuba City FSA Camp, California, on August 18, 1940. Switch between the original and the same 5.35 seconds restored with Diffio 4.7 Flash, both at −20 LUFS.

A 5.35-second passage from a 1940 disc recording, restored with Diffio 4.7 Flash in the Diffio web app and level-matched. Estimated speaker similarity is 0.75. Machine measurements, not a listening test.

- [Before and after, MP3](https://diffio.ai/media/press/oral-history-audio-restoration_before-after.mp3) 462 KB Restored speech from 0:06.35 (Diffio 4.7 Flash) · Original 0:00 to 0:05.35
- [Original, 5.35 s, WAV](https://diffio.ai/media/press/oral-history-1940_original_5s.wav) 1003 KB

**Transcript**

…out on our back porch, and had it just loaded with everything you could think of.

Credit: Recording: "Interview about the Johnson's wedding day", Yuba City FSA Camp, August 18, 1940. Charles L. Todd and Robert Sonkin migrant workers collection (AFC 1985/001), American Folklife Center, Library of Congress. Restored speech: Diffio 4.7 Flash. No endorsement implied.

### Fix phone call audio for a podcast: before and after

Listen for: At 0:08, the pause before “and the varying levels of protection”; at 0:05, the phone line’s noise under “National Monument”.

Original![Spectrogram of the phone guest's answer as recorded, 0 to 4 kHz, line noise throughout and nothing above the phone band](https://diffio.ai/media/press/previews/phone-call_spectrogram-original_1600x360.webp)

Restored speech![Spectrogram of the same passage restored with Diffio 4.7 Flash, 0 to 4 kHz](https://diffio.ai/media/press/previews/phone-call_spectrogram-restored-diffio-4.7-flash_1600x360.webp)

**Spectrograms, 0 to 4 kHz, 10.25 s, on one dB scale; the line at 0:08 marks the “Listen for” moment.**

[Restored speech, Diffio 4.7 Flash, 10.25 s, WAV](https://diffio.ai/media/press/phone-call-interview_restored-diffio-4.7-flash_10s.wav) 1.9 MB

A NOAA podcast guest on a phone line in 2015. Switch between the original and the same 10.25 seconds restored with Diffio 4.7 Flash, both at −23 LUFS. The phone line carried the voice up to about 3.5 kHz.

A 10.25-second answer from a podcast guest on a phone line, restored with Diffio 4.7 Flash in the Diffio web app and level-matched. Estimated speaker similarity is 0.68 on the full 18-second passage, so check the guest's words against the original before quoting them. Machine measurements, not a listening test.

- [Before and after, MP3](https://diffio.ai/media/press/fix-phone-call-audio-for-podcast_before-after.mp3) 845 KB Restored speech from 0:11.25 (Diffio 4.7 Flash) · Original 0:00 to 0:10.25 Listen for in the MP3: At 0:08 and 0:19, the pause before “and the varying levels of protection”; at 0:05 and 0:16, the phone line’s noise under “National Monument”.
- [Original, 10.25 s, WAV](https://diffio.ai/media/press/phone-call-interview_original_10s.wav) 1.9 MB

**Transcript**

Yeah, I think it's worth noting a key distinction between the Papahānaumokuākea Marine National Monument and the high levels of protection they have and the varying levels of protection…

Credit: Audio: "Diving Deeper: Exploring Hawaii's Coral Reefs", NOAA Ocean Service podcast, December 2015, public domain. Excerpted and level-matched. Speech restored with Diffio 4.7 Flash. No endorsement by NOAA or the speakers is implied. [License (opens in a new tab)](https://creativecommons.org/publicdomain/mark/1.0/)

## Before and after graphics

- **Media scrum graphics and strips:** Source: "Ryan Strome, New York Islanders 2011 Draft Pick" by TermiteMound, CC BY 3.0, via Wikimedia Commons. Excerpted, level-matched and speech-restored by Diffio. No endorsement implied. [License (opens in a new tab)](https://creativecommons.org/licenses/by/3.0/)
- **Echo room graphics and strips:** Video: VOA News, "Red Cross Shelter Serves Florence Evacuees" (September 14, 2018), public domain. Speech restored by Diffio.
- **Archived sermon graphics and strips:** Sermon: T. Ernest Wilson, "The God Of All Grace", February 13, 1969, Voices For Christ (vfc.diffio.ai/sermons/52421). Restored by Diffio.
- **1940 oral history graphics and strips:** Recording: "Interview about the Johnson's wedding day", Yuba City FSA Camp, August 18, 1940. Charles L. Todd and Robert Sonkin migrant workers collection (AFC 1985/001), American Folklife Center, Library of Congress. No known restrictions. Restored speech: Diffio 4.7 Flash. No endorsement implied.
- **Flight line strips:** Footage: "Marine Attack Squadron 231 Flight Line Interviews", U.S. Marine Corps video by Cpl. Joshua Munsen (February 13, 2025), via DVIDS. Public domain. Excerpted, markings blurred and speech-restored by Diffio. The appearance of U.S. Department of War (DoW) visual information does not imply or constitute DoW endorsement.
- **Phone call strips:** Audio: "Diving Deeper: Exploring Hawaii's Coral Reefs", NOAA Ocean Service podcast, December 2015, public domain. Excerpted and level-matched. Speech restored with Diffio 4.7 Flash. No endorsement by NOAA or the speakers is implied.

[![Two 0 to 8 kHz spectrograms of an 11.4-second interview answer: grey crowd noise fills gaps between words on top; gaps turn white and dark voice harmonics stand out below.](https://diffio.ai/media/press/previews/remove-crowd-noise-from-interview-audio_spectrogram_1920x1080.webp)](https://diffio.ai/media/press/remove-crowd-noise-from-interview-audio_spectrogram_1920x1080.png)

**Media scrum, annotated Camera audio from a 2011 NHL Draft media scrum (top) and the same 11.4 seconds restored with Diffio 4.7 Flash (bottom), both at −20 LUFS. Between his words the crowd is about 19 dB quieter, and his voice stays within 1 dB of its level. [Hear both](https://diffio.ai/press/#audio-media-scrum-ab) [PNG, 1920 × 1080](https://diffio.ai/media/press/remove-crowd-noise-from-interview-audio_spectrogram_1920x1080.png) 839 KB [JPG, 1200 × 675](https://diffio.ai/media/press/remove-crowd-noise-from-interview-audio_spectrogram_1200x675.jpg) 219 KB Credit: [Media scrum graphics and strips](https://diffio.ai/press/#credit-media-scrum-graphics), at the top of this section**

[![Two 0 to 8 kHz spectrograms of an 11.4-second interview answer: grey crowd noise fills gaps between words on top; gaps turn white and dark voice harmonics stand out below.](https://diffio.ai/media/press/previews/remove-crowd-noise-from-interview-audio_spectrogram-clean_1920x1080.webp)](https://diffio.ai/media/press/remove-crowd-noise-from-interview-audio_spectrogram-clean_1920x1080.png)

**Media scrum, clean Camera audio from a 2011 NHL Draft media scrum (top) and the same 11.4 seconds restored with Diffio 4.7 Flash (bottom), both at −20 LUFS. Between his words the crowd is about 19 dB quieter, and his voice stays within 1 dB of its level. [Hear both](https://diffio.ai/press/#audio-media-scrum-ab) [PNG, 1920 × 1080](https://diffio.ai/media/press/remove-crowd-noise-from-interview-audio_spectrogram-clean_1920x1080.png) 1.0 MB [JPG, 1200 × 675](https://diffio.ai/media/press/remove-crowd-noise-from-interview-audio_spectrogram-clean_1200x675.jpg) 227 KB Credit: [Media scrum graphics and strips](https://diffio.ai/press/#credit-media-scrum-graphics), at the top of this section**

[![Two 0 to 8 kHz spectrograms of a 10.1-second school-cafeteria interview: grey room noise and echo tail fill word gaps on top; gaps clear out and voice harmonics sharpen after Diffio 4.7 Pro below.](https://diffio.ai/media/press/previews/remove-echo-from-interview-audio_spectrogram_1920x1080.webp)](https://diffio.ai/media/press/remove-echo-from-interview-audio_spectrogram_1920x1080.png)

**Echo room, annotated A Red Cross volunteer interviewed by VOA News in a school cafeteria used as a hurricane shelter (top) and the same 10.1 seconds restored with Diffio 4.7 Pro (bottom), both at −20 LUFS. Between his words the level is about 17 dB lower (room noise and echo tail together), and his voice stays within 1 dB of its level. [Hear both](https://diffio.ai/press/#audio-echo-room-ab) [PNG, 1920 × 1080](https://diffio.ai/media/press/remove-echo-from-interview-audio_spectrogram_1920x1080.png) 816 KB [JPG, 1200 × 675](https://diffio.ai/media/press/remove-echo-from-interview-audio_spectrogram_1200x675.jpg) 208 KB Credit: [Echo room graphics and strips](https://diffio.ai/press/#credit-echo-room-graphics), at the top of this section**

[![Two 0 to 8 kHz spectrograms of a 10.1-second school-cafeteria interview: grey room noise and echo tail fill word gaps on top; gaps clear out and voice harmonics sharpen after Diffio 4.7 Pro below.](https://diffio.ai/media/press/previews/remove-echo-from-interview-audio_spectrogram-clean_1920x1080.webp)](https://diffio.ai/media/press/remove-echo-from-interview-audio_spectrogram-clean_1920x1080.png)

**Echo room, clean A Red Cross volunteer interviewed by VOA News in a school cafeteria used as a hurricane shelter (top) and the same 10.1 seconds restored with Diffio 4.7 Pro (bottom), both at −20 LUFS. Between his words the level is about 17 dB lower (room noise and echo tail together), and his voice stays within 1 dB of its level. [Hear both](https://diffio.ai/press/#audio-echo-room-ab) [PNG, 1920 × 1080](https://diffio.ai/media/press/remove-echo-from-interview-audio_spectrogram-clean_1920x1080.png) 1.0 MB [JPG, 1200 × 675](https://diffio.ai/media/press/remove-echo-from-interview-audio_spectrogram-clean_1200x675.jpg) 212 KB Credit: [Echo room graphics and strips](https://diffio.ai/press/#credit-echo-room-graphics), at the top of this section**

[![Two 0 to 3 kHz spectrograms of a 13.4-second 1969 sermon: blocky white holes break up the harmonics, outlined between 3.9 and 6.1 seconds on top; continuous harmonics appear below.](https://diffio.ai/media/press/previews/restore-old-sermon-recording_spectrogram_1920x1080.webp)](https://diffio.ai/media/press/restore-old-sermon-recording_spectrogram_1920x1080.png)

**Archived sermon, annotated T. Ernest Wilson preaching on February 13, 1969, from an archived copy that holds speech up to about 3 kHz (top), and the same 13.4 seconds restored with Diffio 4.7 Flash (bottom), both at −20 LUFS. Machine transcripts of the two agree on all 37 words, but Diffio's restore is guided by a transcript, so that doesn't prove no word changed; estimated speaker similarity is 0.74, so check the voice and words against the archived copy. [Hear both](https://diffio.ai/press/#audio-sermon-ab) [PNG, 1920 × 1080](https://diffio.ai/media/press/restore-old-sermon-recording_spectrogram_1920x1080.png) 726 KB [JPG, 1200 × 675](https://diffio.ai/media/press/restore-old-sermon-recording_spectrogram_1200x675.jpg) 206 KB Credit: [Archived sermon graphics and strips](https://diffio.ai/press/#credit-sermon-graphics), at the top of this section**

[![Two 0 to 3 kHz spectrograms of a 13.4-second 1969 sermon: blocky white holes break up the harmonics between 3.9 and 6.1 seconds on top; continuous harmonics appear below.](https://diffio.ai/media/press/previews/restore-old-sermon-recording_spectrogram-clean_1920x1080.webp)](https://diffio.ai/media/press/restore-old-sermon-recording_spectrogram-clean_1920x1080.png)

**Archived sermon, clean T. Ernest Wilson preaching on February 13, 1969, from an archived copy that holds speech up to about 3 kHz (top), and the same 13.4 seconds restored with Diffio 4.7 Flash (bottom), both at −20 LUFS. Machine transcripts of the two agree on all 37 words, but Diffio's restore is guided by a transcript, so that doesn't prove no word changed; estimated speaker similarity is 0.74, so check the voice and words against the archived copy. [Hear both](https://diffio.ai/press/#audio-sermon-ab) [PNG, 1920 × 1080](https://diffio.ai/media/press/restore-old-sermon-recording_spectrogram-clean_1920x1080.png) 899 KB [JPG, 1200 × 675](https://diffio.ai/media/press/restore-old-sermon-recording_spectrogram-clean_1200x675.jpg) 206 KB Credit: [Archived sermon graphics and strips](https://diffio.ai/press/#credit-sermon-graphics), at the top of this section**

[![Two spectrograms of the same 5.35 seconds of a 1940 interview, 0 to 4 kHz. Top, the original disc: grey hiss fills the gaps and blurs the harmonics. Bottom, the restored speech: white gaps and sharper harmonic bands.](https://diffio.ai/media/press/previews/oral-history-audio-restoration-1940_spectrogram_1920x1080.webp)](https://diffio.ai/media/press/oral-history-audio-restoration-1940_spectrogram_1920x1080.png)

**1940 oral history, annotated An interview recorded on disc at the Yuba City FSA Camp on August 18, 1940 (top), and the same 5.35 seconds restored with Diffio 4.7 Flash (bottom), both at −20 LUFS, on one dB scale, plotted 0 to 4 kHz, the band the original holds. Estimated speaker similarity 0.75; a machine transcript of the restored speech matches the caption transcript. Machine measurements, not a listening test. [Hear both](https://diffio.ai/press/#audio-oral-history-ab) [PNG, 1920 × 1080](https://diffio.ai/media/press/oral-history-audio-restoration-1940_spectrogram_1920x1080.png) 645 KB [JPG, 1200 × 675](https://diffio.ai/media/press/oral-history-audio-restoration-1940_spectrogram_1200x675.jpg) 198 KB Credit: [1940 oral history graphics and strips](https://diffio.ai/press/#credit-oral-history-graphics), at the top of this section**

[![Two spectrograms of the same 5.35 seconds of a 1940 interview, 0 to 4 kHz. Top, the original disc: grey hiss fills the gaps and blurs the harmonics. Bottom, the restored speech: white gaps and sharper harmonic bands.](https://diffio.ai/media/press/previews/oral-history-audio-restoration-1940_spectrogram-clean_1920x1080.webp)](https://diffio.ai/media/press/oral-history-audio-restoration-1940_spectrogram-clean_1920x1080.png)

**1940 oral history, clean An interview recorded on disc at the Yuba City FSA Camp on August 18, 1940 (top), and the same 5.35 seconds restored with Diffio 4.7 Flash (bottom), both at −20 LUFS, on one dB scale, plotted 0 to 4 kHz, the band the original holds. Estimated speaker similarity 0.75; a machine transcript of the restored speech matches the caption transcript. Machine measurements, not a listening test. [Hear both](https://diffio.ai/press/#audio-oral-history-ab) [PNG, 1920 × 1080](https://diffio.ai/media/press/oral-history-audio-restoration-1940_spectrogram-clean_1920x1080.png) 862 KB [JPG, 1200 × 675](https://diffio.ai/media/press/oral-history-audio-restoration-1940_spectrogram-clean_1200x675.jpg) 210 KB Credit: [1940 oral history graphics and strips](https://diffio.ai/press/#credit-oral-history-graphics), at the top of this section**

[![Three-step diagram headed "Mic noise reduction during recording; Diffio afterwards": 1, mic capture, 1 baked track with decisions locked in; 2, Diffio splitting restored speech and background afterwards; 3, side-by-side quote check and Background amount slider.](https://diffio.ai/media/press/previews/wireless-mic-noise-reduction-vs-ai-audio-restoration_1920x1080.webp)](https://diffio.ai/media/press/wireless-mic-noise-reduction-vs-ai-audio-restoration_1920x1080.png)

**Wireless mic noise reduction vs speech restoration after recording Wireless mic noise reduction (such as Basic or Strong on a DJI Mic Mini) works while you record; Diffio works afterwards on the file you already have, outputting restored speech beside a background track while keeping the original in the project with a Restored toggle and Background amount fader. On a live call, a real-time filter such as Krisp is the tool. [PNG, 1920 × 1080](https://diffio.ai/media/press/wireless-mic-noise-reduction-vs-ai-audio-restoration_1920x1080.png) 233 KB [JPG, 1200 × 675](https://diffio.ai/media/press/wireless-mic-noise-reduction-vs-ai-audio-restoration_1200x675.jpg) 145 KB Credit: DJI Mic Mini noise reduction levels (Basic, Strong): DJI; Newsshooter. Spectrograms: Source: "Ryan Strome, New York Islanders 2011 Draft Pick" by TermiteMound, CC BY 3.0, via Wikimedia Commons. Excerpted, level-matched and speech-restored by Diffio. No endorsement implied. [License (opens in a new tab)](https://creativecommons.org/licenses/by/3.0/)**

**The diagram as text**

| Step | Who | What it does |
| --- | --- | --- |
| 1. During capture | Your mic | A mic close to the speaker picks up less of the room. Some mics also reduce noise while recording; the DJI Mic Mini has 2 levels (Basic and Strong), chosen before you start |
| 2. After recording | Diffio | Works on the audio or video file you already have. Separates the restored speech from the background: removes background noise, reduces echo and repairs supported clipping |
| 3. Before you publish | Your edit | Hear the original and restored speech side by side, set how much background to keep, check quotes against the original, then export |
| On a live call | A real-time filter such as Krisp | Works during the call; Diffio isn't a live filter |

[![Scatter plot of 7 speech cleanup runs on 59 recordings: Diffio, Adobe and Cleanvoice sit at +1.87 to +2.34 predicted SCOREQ gain and 0.59 to 0.65 estimated similarity; Auphonic, ElevenLabs and Descript at +0.46 to +0.82 and 0.77 to 0.84.](https://diffio.ai/media/press/previews/ai-speech-enhancer-benchmark_scoreq-vs-speaker-similarity.webp)](https://diffio.ai/media/press/ai-speech-enhancer-benchmark_scoreq-vs-speaker-similarity.png)

**Speech enhancer test, quality gain vs estimated speaker similarity Mean model-predicted speech-quality gain (SCOREQ) against mean estimated speaker similarity, for 7 speech cleanup runs across 59 recordings, with 95% bootstrap intervals. Tools that remove more read lower: an untouched input scored against itself reads about 0.92 (dashed line), so the estimate measures how close the voice stays to the input, not how clean it is. Diffio ran the test; both scores are model predictions, not a human listening test. Full results and Diffio's losses: https://diffio.ai/speech-enhancer-benchmark/. [PNG, 2400 × 1480](https://diffio.ai/media/press/ai-speech-enhancer-benchmark_scoreq-vs-speaker-similarity.png) 292 KB Credit: Chart: Diffio, from its October 2026 speech enhancer test (Diffio 4.7 retested November 3, 2026).**

[![Spectrogram of the scrum answer as recorded, 0 to 8 kHz, crowd noise throughout.](https://diffio.ai/media/press/previews/media-scrum_spectrogram-original_1600x360.webp)](https://diffio.ai/media/press/media-scrum_spectrogram-original_1600x360.png)

**Raw strip: scrum original [PNG, 1600 × 360](https://diffio.ai/media/press/media-scrum_spectrogram-original_1600x360.png) 673 KB Credit: [Media scrum graphics and strips](https://diffio.ai/press/#credit-media-scrum-graphics), at the top of this section**

[![Spectrogram of the same answer restored with Diffio 4.7 Flash, 0 to 8 kHz.](https://diffio.ai/media/press/previews/media-scrum_spectrogram-restored-diffio-4.7-flash_1600x360.webp)](https://diffio.ai/media/press/media-scrum_spectrogram-restored-diffio-4.7-flash_1600x360.png)

**Raw strip: scrum restored [PNG, 1600 × 360](https://diffio.ai/media/press/media-scrum_spectrogram-restored-diffio-4.7-flash_1600x360.png) 644 KB Credit: [Media scrum graphics and strips](https://diffio.ai/press/#credit-media-scrum-graphics), at the top of this section**

[![Spectrogram of the archived sermon passage, 0 to 3 kHz.](https://diffio.ai/media/press/previews/old-sermon_spectrogram-original_1600x360.webp)](https://diffio.ai/media/press/old-sermon_spectrogram-original_1600x360.png)

**Raw strip: sermon original [PNG, 1600 × 360](https://diffio.ai/media/press/old-sermon_spectrogram-original_1600x360.png) 476 KB Credit: [Archived sermon graphics and strips](https://diffio.ai/press/#credit-sermon-graphics), at the top of this section**

[![Spectrogram of the same passage restored with Diffio 4.7 Flash, 0 to 3 kHz.](https://diffio.ai/media/press/previews/old-sermon_spectrogram-restored-diffio-4.7-flash_1600x360.webp)](https://diffio.ai/media/press/old-sermon_spectrogram-restored-diffio-4.7-flash_1600x360.png)

**Raw strip: sermon restored [PNG, 1600 × 360](https://diffio.ai/media/press/old-sermon_spectrogram-restored-diffio-4.7-flash_1600x360.png) 498 KB Credit: [Archived sermon graphics and strips](https://diffio.ai/press/#credit-sermon-graphics), at the top of this section**

[![Spectrogram of the echo-room answer as recorded in a school cafeteria, 0 to 8 kHz, grey room noise and echo tail between words.](https://diffio.ai/media/press/previews/echo-room_spectrogram-original_1600x360.webp)](https://diffio.ai/media/press/echo-room_spectrogram-original_1600x360.png)

**Echo room strip, original [PNG, 1600 × 360](https://diffio.ai/media/press/echo-room_spectrogram-original_1600x360.png) 609 KB Credit: [Echo room graphics and strips](https://diffio.ai/press/#credit-echo-room-graphics), at the top of this section**

[![Spectrogram of the same answer restored with Diffio 4.7 Pro, 0 to 8 kHz.](https://diffio.ai/media/press/previews/echo-room_spectrogram-restored-diffio-4.7-pro_1600x360.webp)](https://diffio.ai/media/press/echo-room_spectrogram-restored-diffio-4.7-pro_1600x360.png)

**Echo room strip, restored [PNG, 1600 × 360](https://diffio.ai/media/press/echo-room_spectrogram-restored-diffio-4.7-pro_1600x360.png) 601 KB Credit: [Echo room graphics and strips](https://diffio.ai/press/#credit-echo-room-graphics), at the top of this section**

[![Spectrogram of the flight line answer as recorded, 0 to 8 kHz, jet and ground noise throughout](https://diffio.ai/media/press/previews/flight-line_spectrogram-original_1600x360.webp)](https://diffio.ai/media/press/flight-line_spectrogram-original_1600x360.png)

**Raw strip: flight line original [PNG, 1600 × 360](https://diffio.ai/media/press/flight-line_spectrogram-original_1600x360.png) 651 KB Credit: [Flight line strips](https://diffio.ai/press/#credit-flight-line-strips), at the top of this section**

[![Spectrogram of the same passage restored with Diffio 4.7 Pro, 0 to 8 kHz](https://diffio.ai/media/press/previews/flight-line_spectrogram-restored-diffio-4.7-pro_1600x360.webp)](https://diffio.ai/media/press/flight-line_spectrogram-restored-diffio-4.7-pro_1600x360.png)

**Raw strip: flight line restored [PNG, 1600 × 360](https://diffio.ai/media/press/flight-line_spectrogram-restored-diffio-4.7-pro_1600x360.png) 603 KB Credit: [Flight line strips](https://diffio.ai/press/#credit-flight-line-strips), at the top of this section**

[![Spectrogram of the phone guest's answer as recorded, 0 to 4 kHz, line noise throughout and nothing above the phone band](https://diffio.ai/media/press/previews/phone-call_spectrogram-original_1600x360.webp)](https://diffio.ai/media/press/phone-call_spectrogram-original_1600x360.png)

**Raw strip: phone call original [PNG, 1600 × 360](https://diffio.ai/media/press/phone-call_spectrogram-original_1600x360.png) 577 KB Credit: [Phone call strips](https://diffio.ai/press/#credit-phone-call-strips), at the top of this section**

[![Spectrogram of the same passage restored with Diffio 4.7 Flash, 0 to 4 kHz](https://diffio.ai/media/press/previews/phone-call_spectrogram-restored-diffio-4.7-flash_1600x360.webp)](https://diffio.ai/media/press/phone-call_spectrogram-restored-diffio-4.7-flash_1600x360.png)

**Raw strip: phone call restored [PNG, 1600 × 360](https://diffio.ai/media/press/phone-call_spectrogram-restored-diffio-4.7-flash_1600x360.png) 562 KB Credit: [Phone call strips](https://diffio.ai/press/#credit-phone-call-strips), at the top of this section**

[![Spectrogram of the 1940 interview as recorded, 0 to 4 kHz, disc hiss throughout](https://diffio.ai/media/press/previews/oral-history-1940_spectrogram-original_1600x360.webp)](https://diffio.ai/media/press/oral-history-1940_spectrogram-original_1600x360.png)

**Raw strip: 1940 oral history original [PNG, 1600 × 360](https://diffio.ai/media/press/oral-history-1940_spectrogram-original_1600x360.png) 499 KB Credit: [1940 oral history graphics and strips](https://diffio.ai/press/#credit-oral-history-graphics), at the top of this section**

[![Spectrogram of the same passage restored with Diffio 4.7 Flash, 0 to 4 kHz](https://diffio.ai/media/press/previews/oral-history-1940_spectrogram-restored-diffio-4.7-flash_1600x360.webp)](https://diffio.ai/media/press/oral-history-1940_spectrogram-restored-diffio-4.7-flash_1600x360.png)

**Raw strip: 1940 oral history restored [PNG, 1600 × 360](https://diffio.ai/media/press/oral-history-1940_spectrogram-restored-diffio-4.7-flash_1600x360.png) 463 KB Credit: [1940 oral history graphics and strips](https://diffio.ai/press/#credit-oral-history-graphics), at the top of this section**

## Social cards

- **Media scrum card and the Open Graph card:** Source: "Ryan Strome, New York Islanders 2011 Draft Pick" by TermiteMound, CC BY 3.0, via Wikimedia Commons. Excerpted, level-matched and speech-restored by Diffio. No endorsement implied. [License (opens in a new tab)](https://creativecommons.org/licenses/by/3.0/)

[![Black-and-white photo of a hockey draft pick at a press table beside Original and Restored speech spectrograms, titled "Crowd noise between the words, turned down", with the line "The crowd between his words is about 19 dB quieter".](https://diffio.ai/media/press/previews/remove-crowd-noise-from-interview-audio_card_1200x630.webp)](https://diffio.ai/media/press/remove-crowd-noise-from-interview-audio_card_1200x630.png)

**Media scrum card A 2011 NHL Draft media scrum on camera audio, then the same 11.4 seconds restored with Diffio 4.7 Flash. The crowd between his words is about 19 dB quieter. [Hear both](https://diffio.ai/press/#audio-media-scrum-ab) [Link card, 1200 × 630](https://diffio.ai/media/press/remove-crowd-noise-from-interview-audio_card_1200x630.png) 376 KB [Square, 1080 × 1080](https://diffio.ai/media/press/remove-crowd-noise-from-interview-audio_card_1080x1080.png) 626 KB Credit: [Media scrum card and the Open Graph card](https://diffio.ai/press/#credit-media-scrum-cards), at the top of this section**

[![Black-and-white photo of a Red Cross volunteer in a school cafeteria beside Original and Restored speech spectrograms, titled "Noise and echo between the words, turned down", with the line "About 17 dB lower between his words".](https://diffio.ai/media/press/previews/remove-echo-from-interview-audio_card_1200x630.webp)](https://diffio.ai/media/press/remove-echo-from-interview-audio_card_1200x630.png)

**Echo room card A VOA News interview in an echoing school cafeteria, then the same 10.1 seconds restored with Diffio 4.7 Pro (0 to 8 kHz). Between his words the level is about 17 dB lower (room noise and echo tail together), and his voice stays within 1 dB of its level. [Hear both](https://diffio.ai/press/#audio-echo-room-ab) [Link card, 1200 × 630](https://diffio.ai/media/press/remove-echo-from-interview-audio_card_1200x630.png) 350 KB [Square, 1080 × 1080](https://diffio.ai/media/press/remove-echo-from-interview-audio_card_1080x1080.png) 574 KB Credit: Video: VOA News, "Red Cross Shelter Serves Florence Evacuees" (September 14, 2018), public domain. Speech restored by Diffio.**

[![Black-and-white photo of a Marine in a flight suit on a flight line beside Original and Restored speech spectrograms, titled "Flight line noise under an interview, turned down", with the line "Same 10.5 seconds, both at the same loudness".](https://diffio.ai/media/press/previews/remove-background-noise-from-outdoor-interview_card_1200x630.webp)](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview_card_1200x630.png)

**Flight line card A Marine interviewed on a flight line, then the same 10.5 seconds restored with Diffio 4.7 Pro, both at the same loudness. [Hear both](https://diffio.ai/press/#audio-flight-line-ab) [Link card, 1200 × 630](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview_card_1200x630.png) 363 KB [Square, 1080 × 1080](https://diffio.ai/media/press/remove-background-noise-from-outdoor-interview_card_1080x1080.png) 588 KB Credit: Footage: "Marine Attack Squadron 231 Flight Line Interviews", U.S. Marine Corps video by Cpl. Joshua Munsen (February 13, 2025), via DVIDS. Public domain. Excerpted, markings blurred and speech-restored by Diffio. The appearance of U.S. Department of War (DoW) visual information does not imply or constitute DoW endorsement.**

[![Black panel reading "1969, February 13" beside 0 to 3 kHz sermon spectrograms labelled Original and Restored speech, titled "A 1969 sermon, before and after", with "Estimated speaker similarity 0.74 (1 = identical voice). Compare it with the archived copy."](https://diffio.ai/media/press/previews/restore-old-sermon-recording_card_1200x630.webp)](https://diffio.ai/media/press/restore-old-sermon-recording_card_1200x630.png)

**Archived sermon card T. Ernest Wilson preaching in 1969, from the archived copy, then restored with Diffio 4.7 Flash. Estimated speaker similarity is 0.74, measured on the 30 seconds around the excerpt; compare the voice and words with the archived copy. [Hear both](https://diffio.ai/press/#audio-sermon-ab) [Link card, 1200 × 630](https://diffio.ai/media/press/restore-old-sermon-recording_card_1200x630.png) 245 KB [Square, 1080 × 1080](https://diffio.ai/media/press/restore-old-sermon-recording_card_1080x1080.png) 369 KB Credit: Sermon: T. Ernest Wilson, "The God Of All Grace", February 13, 1969, Voices For Christ (vfc.diffio.ai/sermons/52421). Restored by Diffio.**

[![Diffio AI logo on black with "Speech restoration" and "Restores speech in noisy, echoey and archival recordings, after the fact.", beside Original and Restored speech spectrograms of the 2011 NHL Draft media scrum.](https://diffio.ai/media/press/previews/diffio-ai_open-graph_1200x630.webp)](https://diffio.ai/media/press/diffio-ai_open-graph_1200x630.png)

**Open Graph card [PNG, 1200 × 630](https://diffio.ai/media/press/diffio-ai_open-graph_1200x630.png) 203 KB Credit: [Media scrum card and the Open Graph card](https://diffio.ai/press/#credit-media-scrum-cards), at the top of this section**

## Product screenshots

- **Media scrum project screenshots:** Source: "Ryan Strome, New York Islanders 2011 Draft Pick" by TermiteMound, CC BY 3.0, via Wikimedia Commons. Excerpted, level-matched and speech-restored by Diffio. No endorsement implied. [License (opens in a new tab)](https://creativecommons.org/licenses/by/3.0/)
- **Sermon walkthrough, steps 3 to 5:** Image: Diffio. Recording: T. Ernest Wilson, "The God Of All Grace", February 13, 1969, Voices For Christ (vfc.diffio.ai/sermons/52421).

[![The Diffio player on a flight line interview video at 0:17 of 0:29 with Restored on and Background amount at 30%, beside an enlarged view headed "Restored on, background at 30%".](https://diffio.ai/media/press/previews/diffio-speech-restoration-player_1920x1080.webp)](https://diffio.ai/media/press/diffio-speech-restoration-player_1920x1080.png)

**Player Diffio gives two outputs, restored speech plus a background track you mix back in: here the Restored toggle is on and the Background amount slider is at 30% while a flight line interview plays (this project's Diffio 4.7 Flash result; the flight line videos and audio use Pro; the inset enlarges both controls). Diffio app, November 2026. [PNG, 1920 × 1080](https://diffio.ai/media/press/diffio-speech-restoration-player_1920x1080.png) 428 KB [JPG, 1200 × 675](https://diffio.ai/media/press/diffio-speech-restoration-player_1200x675.jpg) 99 KB Credit: Footage: "Marine Attack Squadron 231 Flight Line Interviews", U.S. Marine Corps video by Cpl. Joshua Munsen (February 13, 2025), via DVIDS. Public domain. Excerpted, markings blurred and speech-restored by Diffio. The appearance of U.S. Department of War (DoW) visual information does not imply or constitute DoW endorsement.**

[![The Diffio player playing a VOA interview with a Red Cross volunteer in a school cafeteria, at 0:17 of 0:35: the Pause button, the Restored toggle on, the waveform and the Background amount slider at 0%.](https://diffio.ai/media/press/previews/diffio-player-playing-restored_2400x2310.webp)](https://diffio.ai/media/press/diffio-player-playing-restored_2400x2310.png)

**The player while it plays The Original and Restored toggle works during playback: the Pause button shows, Restored is on, and the Background amount slider (0% here) sets how much of the room to keep. The player holds this VOA interview's Diffio 4.7 Pro result. Captured November 13, 2026 on the Diffio web app. [PNG, 2400 × 2310](https://diffio.ai/media/press/diffio-player-playing-restored_2400x2310.png) 2.0 MB [JPG, 1200 × 1155](https://diffio.ai/media/press/diffio-player-playing-restored_1200x1155.jpg) 194 KB Credit: Video: VOA News, "Red Cross Shelter Serves Florence Evacuees" (September 14, 2018), public domain. Speech restored by Diffio.**

[![The top of Diffio's Create new project dialog: the file media-scrum-camera-audio.mp4 (video, MP4, 21.7 MB) attached and the project name Media scrum (camera audio).](https://diffio.ai/media/press/previews/diffio-new-project-video-file_1168x634.webp)](https://diffio.ai/media/press/diffio-new-project-video-file_1168x634.png)

**New project, video file Uploading a video file: the top of the Create new project dialog, with the media scrum MP4 attached and a project name. The model choice (Diffio 4.7 Flash was selected) and the Create project button are below this crop. Captured November 13, 2026 on the Diffio web app. [PNG, 1168 × 634](https://diffio.ai/media/press/diffio-new-project-video-file_1168x634.png) 44 KB**

[![A finished Diffio project titled Media scrum (camera audio): the video paused at 1:08 of 1:18, waveform, Restored toggle, Background amount at 0%, Download button and the transcript viewer.](https://diffio.ai/media/press/previews/diffio-project-page_2880x3170.webp)](https://diffio.ai/media/press/diffio-project-page_2880x3170.png)

**Project page One finished project holds the video, the Original and Restored toggle, the Background amount slider (0% here), the Download button and the transcript. Captured November 12, 2026. [PNG, 2880 × 3170](https://diffio.ai/media/press/diffio-project-page_2880x3170.png) 1.2 MB Credit: [Media scrum project screenshots](https://diffio.ai/press/#credit-media-scrum-screenshots), at the top of this section**

[![The whole project page, down to the project details and the results list showing Diffio 4.7 Flash.](https://diffio.ai/media/press/previews/diffio-project-page-full_2880x3720.webp)](https://diffio.ai/media/press/diffio-project-page-full_2880x3720.png)

**Full project page The whole project page, from the player and transcript down to the results list, where this project's Diffio 4.7 Flash result sits. Captured November 12, 2026. [PNG, 2880 × 3720](https://diffio.ai/media/press/diffio-project-page-full_2880x3720.png) 981 KB Credit: [Media scrum project screenshots](https://diffio.ai/press/#credit-media-scrum-screenshots), at the top of this section**

[![The transcript viewer with two paragraphs of the scrum, play, Restored, copy and export buttons.](https://diffio.ai/media/press/previews/diffio-transcript-viewer_2400x950.webp)](https://diffio.ai/media/press/diffio-transcript-viewer_2400x950.png)

**Transcript viewer Every result comes with a transcript you can play from, copy or export. This one is of the uploaded recording, so it includes the reporters' questions. [PNG, 2400 × 950](https://diffio.ai/media/press/diffio-transcript-viewer_2400x950.png) 138 KB Credit: [Media scrum project screenshots](https://diffio.ai/press/#credit-media-scrum-screenshots), at the top of this section**

[![The top of Diffio's Create new project dialog with the file The God Of All Grace - T. Ernest Wilson 1969.flac (audio, FLAC, 562.8 KB) attached and the project name The God Of All Grace (1969).](https://diffio.ai/media/press/previews/clean-up-sermon-audio_step-1_new-project.webp)](https://diffio.ai/media/press/clean-up-sermon-audio_step-1_new-project.png)

**Sermon walkthrough, step 1: new project Step 1. Attach the file and name the project: the top of the Create new project dialog with the sermon file attached and the project name "The God Of All Grace (1969)". The model choice is below this crop; this walkthrough used Diffio 4.7 Pro. Captured November 13, 2026. [PNG, 1168 × 682](https://diffio.ai/media/press/clean-up-sermon-audio_step-1_new-project.png) 51 KB Credit: Image: Diffio.**

[![The Diffio audio player playing at 0:10 of 0:35 with the toggle set to Original, showing a waveform, the Pause button and the Background amount slider at 0%.](https://diffio.ai/media/press/previews/clean-up-sermon-audio_step-3_original.webp)](https://diffio.ai/media/press/clean-up-sermon-audio_step-3_original.png)

**Sermon walkthrough, step 3: original Step 3. The finished project's player on Original while it plays, 10 seconds into the sermon: the waveform, the Pause button, the Original and Restored toggle and the Background amount slider at 0%. Recaptured during playback November 13, 2026. [PNG, 2400 × 796](https://diffio.ai/media/press/clean-up-sermon-audio_step-3_original.png) 99 KB [JPG, 1200 × 398](https://diffio.ai/media/press/clean-up-sermon-audio_step-3_original.jpg) 40 KB Credit: [Sermon walkthrough, steps 3 to 5](https://diffio.ai/press/#credit-sermon-steps), at the top of this section**

[![The Diffio audio player playing at the same 0:10 mark, the Pause button showing, with the toggle switched to Restored and the Background amount slider at 0%.](https://diffio.ai/media/press/previews/clean-up-sermon-audio_step-3_restored.webp)](https://diffio.ai/media/press/clean-up-sermon-audio_step-3_restored.png)

**Sermon walkthrough, step 3: restored Step 3. The same moment with the toggle switched to Restored (Diffio 4.7 Pro): one player flips between the original and the restored speech, so you compare the two at the same point. [PNG, 2400 × 796](https://diffio.ai/media/press/clean-up-sermon-audio_step-3_restored.png) 99 KB [JPG, 1200 × 398](https://diffio.ai/media/press/clean-up-sermon-audio_step-3_restored.jpg) 41 KB Credit: [Sermon walkthrough, steps 3 to 5](https://diffio.ai/press/#credit-sermon-steps), at the top of this section**

[![The Diffio audio player playing on Restored at 0:10, the Pause button showing, with the Background amount slider set to 40%.](https://diffio.ai/media/press/previews/clean-up-sermon-audio_step-4_background-amount.webp)](https://diffio.ai/media/press/clean-up-sermon-audio_step-4_background-amount.png)

**Sermon walkthrough, step 4: Background amount Step 4. The Background amount slider brings part of the background back under the restored speech, here 40%, while it plays. [PNG, 2400 × 796](https://diffio.ai/media/press/clean-up-sermon-audio_step-4_background-amount.png) 99 KB [JPG, 1200 × 398](https://diffio.ai/media/press/clean-up-sermon-audio_step-4_background-amount.jpg) 41 KB Credit: [Sermon walkthrough, steps 3 to 5](https://diffio.ai/press/#credit-sermon-steps), at the top of this section**

[![Seven lines of the sermon transcript in the Diffio app, with the spoken word "forest" highlighted in a grey box.](https://diffio.ai/media/press/previews/clean-up-sermon-audio_step-5_transcript.webp)](https://diffio.ai/media/press/clean-up-sermon-audio_step-5_transcript.png)

**Sermon walkthrough, step 5: transcript Step 5. The transcript of the excerpt, with the word being spoken ("forest") highlighted as the restored speech plays. [PNG, 1436 × 508](https://diffio.ai/media/press/clean-up-sermon-audio_step-5_transcript.png) 92 KB [JPG, 1200 × 425](https://diffio.ai/media/press/clean-up-sermon-audio_step-5_transcript.jpg) 111 KB Credit: [Sermon walkthrough, steps 3 to 5](https://diffio.ai/press/#credit-sermon-steps), at the top of this section**

[![The Download audio dialog showing Restored speech (selected), Current mix and Background noise radio buttons, WAV format selected, Boost quiet speech and the file name the-god-of-all-grace_1969_diffio_speech.wav.](https://diffio.ai/media/press/previews/clean-up-sermon-audio_step-6_download.webp)](https://diffio.ai/media/press/clean-up-sermon-audio_step-6_download.png)

**Sermon walkthrough, step 6: download Step 6. The Download audio dialog offers Restored speech, Current mix (here restored speech with 40% background) or Background noise (the background track alone), in the format you pick: WAV here; the web app exports WAV, MP3 or M4A, plus MP4 for video. [PNG, 920 × 1448](https://diffio.ai/media/press/clean-up-sermon-audio_step-6_download.png) 115 KB Credit: Image: Diffio.**

[![Diffio's Basic plan ($8 a month for 6 hours) and Pro plan ($29 a month for 30 hours), each including Diffio 4.7 Flash, 4.7 Pro and full downloads.](https://diffio.ai/media/press/previews/diffio-pricing-basic-and-pro-plans_1440x1080.webp)](https://diffio.ai/media/press/diffio-pricing-basic-and-pro-plans_1440x1080.png)

**Basic and Pro plans Diffio's paid plans on November 12, 2026; there's also a Free plan ($0 for 3 uploads and a 2-minute Flash preview). Check diffio.ai/pricing before quoting a price. [PNG, 1440 × 1080](https://diffio.ai/media/press/diffio-pricing-basic-and-pro-plans_1440x1080.png) 141 KB [Whole pricing page, PNG, 2880 × 1800](https://diffio.ai/media/press/diffio-pricing_2880x1800.png) 307 KB**

[![Diffio's API page: "Speech restoration API: build Diffio into your product", from $5 an hour billed per second with a $5 starter credit, beside a Python SDK example calling restore_audio with diffio-4.7-flash.](https://diffio.ai/media/press/previews/speech-restoration-api_2880x1592.webp)](https://diffio.ai/media/press/speech-restoration-api_2880x1592.png)

**API page The Diffio API overview on November 13, 2026: the Python and Node SDK example (shown in Python) and the REST base URL https://api.diffio.ai/v2. The example uses diffio 1.0 and API v2, as the current docs do; pip install diffio installs diffio 1.0 and npm install diffio installs 1.0.x. [PNG, 2880 × 1592](https://diffio.ai/media/press/speech-restoration-api_2880x1592.png) 283 KB**

[![The "Clean up interview audio" page: an upload box beside a player with 4 interviews (windy beach, train, media scrum, Skype call) and Original, 4.7 Flash and 4.7 Pro buttons.](https://diffio.ai/media/press/previews/clean-up-interview-audio_2880x1800.webp)](https://diffio.ai/media/press/clean-up-interview-audio_2880x1800.png)

**Interview cleanup page Diffio's interview cleanup page on November 11, 2026. [PNG, 2880 × 1800](https://diffio.ai/media/press/clean-up-interview-audio_2880x1800.png) 275 KB**

[![A Diffio project restoring: "Restoring your audio", "Saving your files…", a nearly full progress bar reading 1:18 of 1:18 restored, and a Results row with Diffio 4.7 Flash marked Processing.](https://diffio.ai/media/press/previews/diffio-restoration-in-progress_2352x667.webp)](https://diffio.ai/media/press/diffio-restoration-in-progress_2352x667.png)

**Restoring A project restoring. You can close the page and the project updates when it finishes. Captured November 13, 2026; the Download button, transcript box and details between the two cards are cropped out. [PNG, 2352 × 667](https://diffio.ai/media/press/diffio-restoration-in-progress_2352x667.png) 58 KB Credit: [Media scrum project screenshots](https://diffio.ai/press/#credit-media-scrum-screenshots), at the top of this section**

## Logos

[![Diffio AI logo.](https://diffio.ai/media/press/previews/diffio-logo_black-on-white.webp)](https://diffio.ai/media/press/diffio-logo_black-on-white.png)

**Wordmark, black on white [PNG, 4461 × 3175](https://diffio.ai/media/press/diffio-logo_black-on-white.png) 45 KB**

[![Diffio AI logo.](https://diffio.ai/media/press/previews/diffio-logo_black_transparent.webp)](https://diffio.ai/media/press/diffio-logo_black_transparent.png)

**Wordmark, black on transparent [PNG, 4461 × 3175](https://diffio.ai/media/press/diffio-logo_black_transparent.png) 85 KB**

[![Diffio AI logo.](https://diffio.ai/media/press/previews/diffio-logo_white-on-black.webp)](https://diffio.ai/media/press/diffio-logo_white-on-black.png)

**Wordmark, white on black [PNG, 4461 × 3175](https://diffio.ai/media/press/diffio-logo_white-on-black.png) 39 KB**

[![Diffio AI logo.](https://diffio.ai/media/press/previews/diffio-logo_white_transparent.webp)](https://diffio.ai/media/press/diffio-logo_white_transparent.png)

**Wordmark, white on transparent [PNG, 4461 × 3175](https://diffio.ai/media/press/diffio-logo_white_transparent.png) 89 KB**

[![Diffio AI logo.](https://diffio.ai/media/press/previews/diffio-icon_black-on-white_1024.webp)](https://diffio.ai/media/press/diffio-icon_black-on-white_1024.png)

**Icon, black on white [PNG, 1024 × 1024](https://diffio.ai/media/press/diffio-icon_black-on-white_1024.png) 4 KB**

[![Diffio AI logo.](https://diffio.ai/media/press/previews/diffio-icon_black_transparent_1024.webp)](https://diffio.ai/media/press/diffio-icon_black_transparent_1024.png)

**Icon, black on transparent [PNG, 1024 × 1024](https://diffio.ai/media/press/diffio-icon_black_transparent_1024.png) 4 KB**

[![Diffio AI logo.](https://diffio.ai/media/press/previews/diffio-icon_white-on-black_1024.webp)](https://diffio.ai/media/press/diffio-icon_white-on-black_1024.png)

**Icon, white on black [PNG, 1024 × 1024](https://diffio.ai/media/press/diffio-icon_white-on-black_1024.png) 2 KB**

[![Diffio AI logo.](https://diffio.ai/media/press/previews/diffio-icon_white_transparent_1024.webp)](https://diffio.ai/media/press/diffio-icon_white_transparent_1024.png)

**Icon, white on transparent [PNG, 1024 × 1024](https://diffio.ai/media/press/diffio-icon_white_transparent_1024.png) 4 KB**

[![Diffio AI logo.](https://diffio.ai/media/press/diffio-logo_source.svg)](https://diffio.ai/media/press/diffio-logo_source.svg)

**Wordmark, vector outlines [SVG](https://diffio.ai/media/press/diffio-logo_source.svg) 37 KB**

## About Diffio AI

Diffio AI restores speech in recordings that can't be made again, like an interview in a loud room or a lecture taped years ago. It removes background noise, reduces echo and repairs some clipping, in audio or video. It runs in a browser, where you hear the original and restored speech side by side before you publish. It's made for podcasters, video producers, newsrooms and educators. Diffio was founded in 2025 by Nathan Harmon in Colorado Springs, Colorado. Learn more at diffio.ai.

Diffio AI is a service provided by Ozni Apps, LLC.

**Other lengths: 10, 25, 50 and 250 words**

10 words

Diffio AI removes noise and reduces echo in recorded speech.

25 words

Diffio AI removes noise and reduces echo in recorded speech. You hear the original and the restored speech side by side before you publish.

50 words

Diffio AI removes background noise, reduces echo and repairs some clipping in recorded speech. It's for podcasters, video producers and newsrooms with recordings that can't be made again, like an interview in a loud room. You hear the original and restored speech side by side. Learn more at diffio.ai.

250 words

Most speech worth keeping was recorded once, in whatever room was there: a reporter's camera inside a crowd, a podcast guest on a phone line, a professor's lectures taped on cassette in the 1980s. The words were good and the room wasn't. There's no second take. Diffio AI restores the speech in the file you already have.

It removes background noise, reduces echo and repairs some clipping, in audio or video, with its own models. You use it in a browser. Upload audio or video up to 90 minutes and hear the original and restored speech side by side. Crowd and music go to a background track, and a slider brings back as much of them as you want. Download the restored speech, or the video with its picture untouched. Developers get the same models through a REST API with Python and Node SDKs.

Diffio was founded in 2025 by Nathan Harmon, who spent 10 years in signal processing for defense and commercial teams, most recently at Northrop Grumman. It's bootstrapped and based in Colorado Springs, Colorado. There's a free plan; paid plans start at $8 per month. Learn more at diffio.ai.

## Fact sheet

Facts as of November 12, 2026.

Company

Diffio AI

Category

Speech restoration

What it does

Restores recorded speech in audio and video: removes background noise, reduces room echo and repairs supported clipping and distortion

Founded

2025

Founder

Nathan Harmon

Headquarters

Colorado Springs, Colorado, USA

Funding

Bootstrapped

Website

diffio.ai · Blog: diffio.ai/blog · X: @diffioai · GitHub: github.com/Diffio-AI

Press contact

Nathan Harmon, nathan.harmon@diffio.ai

Support

support@diffio.ai

### Models

| Detail | Diffio 4.7 Flash | Diffio 4.7 Pro |
| --- | --- | --- |
| Trade-off | Faster | More model work per step, slower; compare both on a hard recording |
| How it works | Less model work per restoration step | More model work per restoration step |
| Speed | About 30× faster than real time | About 10× faster than real time |
| Languages | Best in English; other languages run, with varying quality | Best in English; other languages run, with varying quality |
| Plans | Every plan, including Free | Basic and Pro plans (the Pro model is included in Basic, $8 per month) |
| API model ID | diffio-4.7-flash | diffio-4.7-pro |

**What goes in and what comes out**

Audio

MP3, WAV, M4A, AAC, FLAC, OGG

Video

MP4, MOV, AVI, MKV, WebM

Phone recordings

3GP, 3G2

Per file

in the web app, up to 90 minutes and 2 GB on every plan; through the API, up to 2 GB with no length limit

Two outputs

restored speech and the background. Both the web app (Download dialog: Current mix, Restored speech, Background noise) and the API (mix and exports endpoints) let you save a background mix and export restored speech, the isolated background track or a custom mix in WAV, MP3, M4A or MP4 (for video); the API can also export both tracks in a ZIP. Music and crowd are separated into the background track, not deleted, and the Background amount control mixes back as much as you want.

Video in, same video out

the picture is unchanged and the soundtrack is restored.

Transcripts

in the app, and from the API with timestamps.

**Where it runs**

Web app

in modern browsers (Chrome, Edge, Firefox, Safari on macOS and iOS, Chrome and Samsung Internet on Android)

REST API

(https://api.diffio.ai/v2) with webhooks and a dashboard for keys, usage and billing

SDKs

Python (pip install diffio installs diffio 1.0, which calls API v2) and Node (npm install diffio installs diffio 1.0.x). Both were published November 13, 2026; the current docs use them

**Features**

- Original and restored speech side by side, with a toggle during playback; click the waveform to seek
- Background amount slider, from speech only to keeping the background
- Projects hold the original and every result made from it
- Share projects with a public link (paid plans)

**Plans (web app)**

| Plan | Price | Includes |
| --- | --- | --- |
| Free | $0 | 3 lifetime uploads (audio or video), 2-minute preview of each result, Diffio 4.7 Flash, no download |
| Basic | $8/month | 6 hours (360 minutes) per month, Flash and Pro models, video, download and export |
| Pro | $29/month | 30 hours (1,800 minutes) per month, same features, for professionals and businesses |

**API pricing (separate from the app plans)**

- Flash: $5 per hour of source audio or video (about $0.001389 per second)
- Pro: $10 per hour of source audio or video (about $0.002778 per second)
- A one-time $5 credit to start, no card needed; after that, prepaid credit added $5 to $100 at a time, with optional automatic top-up. Billed per second, minimum 60 billable seconds per generation

**Who it's for**

Podcasters, video producers, newsrooms and field reporters, educators with recurring recordings and developers building speech restoration into their products.

**Customer voice**

> “Back in the 1980s I recorded a series of lectures that had excellent content but only marginal audio quality, on cassette tapes. It would be impossible to reproduce them now with the same energy. The lectures are now being reproduced with near studio quality audio and enjoyed by many who weren't even alive when I first delivered them.”

**Bruce W. Gore, M.A., J.D., retired lawyer and college professor**

**Worked example**

Original

a 2011 NHL Draft media scrum, 78 seconds of camera audio recorded inside the crowd; Diffio 4.7 ran on a 30 s clip of it (TermiteMound, CC BY 3.0)

With Diffio 4.7 Flash

on an 11.4-second answer, the crowd between his words is about 19 dB quieter and his voice stays within 1 dB of its level (both files level-matched to −20 LUFS)

Transcript check

an automated speech recognizer misheard "Niagara IceDogs" on the original and recognized it on the restored file

[Hear both](https://diffio.ai/press/#audio-media-scrum-ab)

### Can Diffio change what someone said?

On a clean recording it shouldn't happen; under heavy noise, where words are hard to make out, it can. In our October 2026 test of 59 clips (Diffio retested November 3, 2026), mean word error rate of a small speech recognizer against a reference transcript of the original was 3.9% for Diffio 4.7 Flash and 4.1% for Diffio 4.7 Pro, with the untouched originals at 13.2% (Auphonic 13.3%, Cleanvoice 16.4%, ElevenLabs Voice Isolator 18.1%, Adobe Podcast Enhance Speech v2 18.2%, Descript Studio Sound 22.6%). Diffio's restoration is guided by that same reference transcript, which favors it. Diffio did not lead in four places:

- **Phone and remote calls:** Cleanvoice had the lower word error rate.
- **Narration over music:** ElevenLabs Voice Isolator had the lower word error rate, and Adobe Podcast the higher quality gain.
- **Echoey rooms:** Cleanvoice and Auphonic had a lower word error rate than Diffio 4.7 Pro.
- **Hum and machine noise:** Adobe's quality gain was level with Diffio 4.7 Flash's.

Check anything that matters against the original beside the result in the player.

## Founder

![Nathan Harmon.](https://diffio.ai/media/press/nathan-harmon_headshot_400x400.png)

### Nathan Harmon

Nathan Harmon founded Diffio AI in 2025 after 10 years in signal processing for defense and commercial teams, most recently RF geolocation at Northrop Grumman. Diffio restores recorded speech, and he still trains the models and runs the machines himself. It's bootstrapped and based in Colorado Springs, Colorado.

[Headshot, PNG, 400 × 400](https://diffio.ai/media/press/nathan-harmon_headshot_400x400.png) 171 KB
