Skip to content
yepher

Training Audio Recorder

A guided session that records your room for training noise-suppression models: room tone, a measurement sweep, background audio through a real speaker, and short spoken answers, each checked as it is recorded. Invite only; recordings are uploaded with your consent.

Uses the serverInvite only
Loading…

What a session looks like

  1. Consent and setup. Read and sign the consent, describe the room, pick the microphone and the speaker the background plays from. The page turns off the browser's echo cancellation, noise suppression and auto-gain, and shows whether that worked.
  2. Room tone (30 s). Silence, to measure the room. If it's too loud the page won't go on.
  3. Sweep. A rising tone at two levels through the background speaker. The page turns it into the room's impulse response and checks it right away. Bluetooth and AirPlay delay doesn't matter.
  4. Background only. Speech and music clips (or your own TV) at a few volumes, with nobody talking.
  5. Talker only. Questions appear on screen and you answer out loud, briefly, like a caller. Make up the answers.
  6. Again, differently. Move the speaker, sit further back or switch mics, and repeat. Each setup takes about four minutes.

Every take starts after a countdown and stops by itself, so nobody has to touch the laptop while it records. Takes upload as you go; when you finish, the session moves to a private Hugging Face dataset and is deleted from yepher.com.