Make audio sound better
Most speech recordings that sound bad have one of a handful of problems: noise, room sound, wind, hum, distortion or a poor mic. Find yours below and fix it.
Or skip the diagnosis. Upload the file to Diffio, which restores the speech in audio and video, and compare the result with the original.
Make a recording sound better
Drag an audio or video file here, orPick an audio or video file from your phone
MP3, WAV, M4A, AAC, FLAC or OGG audio, or MP4, MOV, AVI, MKV or WebM video. Up to 90 minutes and 2 GB per file.
Free plan: 3 uploads, audio or video, with a 2-minute preview of each result. Downloads and the Pro model from $8 per month.
One bad recording, before and after
This clip has four problems at once. Press play on the original, then switch to the restored speech at the same point to hear what changes.
Listen for: Four problems at once: a dull, muffled voice, a mic far from the talker, a lecture-room echo and cafe chatter. Try to follow the words in the original first.
Sample made from clean speech (LibriTTS, CC BY 4.0), dulled with a 3.5 kHz low-pass filter, played through a real measured lecture-room impulse response (BUT Speech@FIT Reverb Database, CC BY 4.0), mixed with a real cafe recording (by mikemoons, Freesound, CC0) and saved 30 dB below full scale. Both sides are loudness-matched for playback, so the low recording level is not audible here. Restored with Diffio 4.6 Pro. Both clips are loudness-matched for playback.
- Speech: LibriTTS test-clean, speaker 4077, LibriTTS (Zen et al., 2019), from a public domain LibriVox reading, CC BY 4.0.
- Room impulse response: BUT Speech@FIT Reverb Database, room D105 (lecture room, about 2,000 m³), Brno University of Technology, Speech@FIT (Szöke et al., 2019), CC BY 4.0.
- Cafe recording: cafe ambience, mikemoons, CC0 1.0.
Each pair is the clip we uploaded and the restored speech Diffio returned on November 3, 2026, with one fixed volume change so both play at the same loudness. Full credits and the sample record are published with the files.
Find what is wrong with the recording
Listen on headphones and match what you hear. Each fix links to a page with its own before and after samples and the manual steps.
| What you hear | Usual cause | Fix |
|---|---|---|
| A steady bed of noise under the voice: air conditioning, traffic, a crowd | A noisy place, or a mic far from the speaker | Remove background noise |
| Rumble and thumps outdoors | Air hitting the mic | Remove wind noise |
| A low buzz, a bright hiss or crackle | Ground loops, preamps, tape, bad cables | Remove hum, hiss and static |
| A hollow, far-away voice with a tail after each word | A hard room and a distant mic | Remove reverb |
| Words repeating a moment later | A far wall or a call feeding back | Remove echo |
| Harsh crackle on the loudest words | Gain set too high, clipping | Fix distorted audio |
| Thin, tinny or muffled speech | A laptop, phone or webcam mic | Fix bad mic audio |
Too quiet is a different problem. Turning up a quiet recording turns up its noise as well, so remove the noise first and set the level afterward with Normalize or Loudness Normalization in your editor.
How to make audio sound better with Diffio
Real recordings rarely have just one problem. The sample above was muffled, recorded from across an echoey room and mixed with cafe chatter. Diffio restores the speech as a whole instead of running one filter per problem.
- Upload the original. MP3, WAV, M4A, AAC, FLAC or OGG audio, or MP4, MOV, AVI, MKV or WebM video, up to 90 minutes and 2 GB on every plan.
- Choose a model. Diffio 4.6 Flash is fast and on every plan, including Free. Diffio 4.6 Pro restores every window 3 times and keeps the cleanest. The Pro model is included in Basic.
- Compare. Switch between the original and the restored speech on the same playhead and check the parts that worried you.
- Download. On a paid plan, download the restored speech or the full video with the picture untouched. Diffio returns the background as its own track, and a slider mixes back as much of it as you want. Every project comes with a transcript too.
Get better audio next time
- Move the mic closer. Halving the distance to the speaker helps more than any setting. A lavalier or a handheld mic a few inches away beats a camera mic across the room.
- Choose a soft, quiet room. Carpet and furniture cut reverb. Switch off fans and air conditioning for the take if you can.
- Set the level with headroom. Aim for the loudest words to stay well below the maximum, so nothing clips.
- Record to WAV or a high bitrate, and keep the original file. Every re-export loses a little.
- Outdoors, use a windscreen, and turn the speaker's back to the wind.
- Listen on headphones while recording. Hum, wind and a loose cable are easy to miss on speakers.
How to improve audio quality in Audacity, Premiere Pro and DaVinci Resolve
Fixing a recording by hand means one tool per problem, in order: noise and hum first, then EQ, then level. Each defect page above has the detailed steps.
Audacity (free)
- Remove steady noise with Effect, Noise Removal and Repair, Noise Reduction: get a noise profile from a stretch with no speech, then apply it to the whole clip.
- Cut rumble with Effect, EQ and Filters, High-Pass Filter at about 80 Hz.
- Set the level last with Normalize or Loudness Normalization.
Adobe Premiere Pro
- Select the clip and tag it as Dialogue in the Essential Sound panel.
- Under Repair, turn on only the problems you hear: Reduce Noise, Reduce Rumble, DeHum or Reduce Reverb.
- Use Clarity for EQ, then set the level with Loudness.
DaVinci Resolve
- Add Noise Reduction or De-Hummer from the Fairlight FX library for steady noise and hum.
- In DaVinci Resolve Studio, the Voice Isolation control in the Inspector separates the voice from background noise.
- Shape the voice with the track EQ on the Fairlight page, then set the level.
What Diffio can't fix
- Missing words. Speech that was never captured, or that noise covered completely, isn't there to restore.
- Overlapping speakers. Two people talking at once stay together in one speech track.
- Music. Diffio restores speech. Music goes to the background track rather than being restored.
- Languages other than English. Diffio works best in English. It runs on other languages, but results aren't as good yet, so try your own clip first.
Questions about improving audio quality
How can I improve audio quality online?
Upload the recording to Diffio in your browser. It removes background noise, reduces echo and repairs some clipping in the speech, and you compare the result with the original before you download anything.
Which is better sound quality, 128k or 256k?
For MP3, 256 kbps keeps more detail than 128 kbps. Converting a 128 kbps file to 256 kbps adds nothing back, though, and bitrate doesn't remove noise or echo. When you have the choice, export at the higher rate and upload the original file.
How do I get the highest quality audio when recording?
Put the mic close to the speaker, record in a small room with soft furnishings, set the level so the loudest words stay well below the maximum and record to WAV or a high bitrate. Outdoors, add a windscreen.
Can AI make a bad recording sound better?
Often, yes, when the problem is noise, room sound, hum or wind on a voice. It can't bring back words that were never captured, so judge the result against the original on your own file.
What can't Diffio fix?
Two people talking at once stay together in one speech track, because Diffio doesn't separate speakers. Speech that is mostly missing, drowned out or heavily clipped improves less. Diffio works best in English.
Will it make music sound better?
No. Diffio restores speech. Music is separated into the background track, so you can mix it back in, but it isn't restored.
Is it free to try?
Yes. The Free plan includes 3 uploads, audio or video, with a 2-minute preview of each result on Diffio 4.6 Flash. Full playback, downloads and the Pro model start with Basic at $8 per month. See plans and limits.
Hear what your recording can sound like
Upload the recording you're least happy with and hear the restored speech next to the original. Results depend on the recording, which is why the first 3 are free.