Audio samples on https://diffio.ai/make-audio-sound-better/

Each "original" is the clip uploaded to Diffio. Each "pro" file is the unedited restored speech that the
Diffio API returned with Diffio 4.6 Pro on November 3, 2026, with no background mixed back in. The only change to
either file is one fixed gain so both sides of a pair play at the same integrated loudness. The restored files
are adaptations of their sources and carry the same license.

[general-02]
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.
- clean speech: LibriTTS test-clean speaker 4077, utterances 4077_13751_000023_000000 to 4077_13751_000023_000005, LibriTTS (Zen et al., 2019), from a public-domain LibriVox reading. https://www.openslr.org/60/ (CC BY 4.0, https://creativecommons.org/licenses/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). https://speech.fit.vut.cz/software/but-speech-fit-reverb-database (CC BY 4.0, https://creativecommons.org/licenses/by/4.0/)
- impairment recording: cafe ambience, mikemoons. https://freesound.org/people/mikemoons/sounds/579500/ (CC0 1.0, https://creativecommons.org/publicdomain/zero/1.0/)
Files: general-02-original.mp3, general-02-pro.mp3