Skip to content

Known-good microphones

Microphone quality affects transcription accuracy more than almost anything else you can change — more than the Whisper model size, in most rooms. This page is a community list: real microphones, on real machines, with the vad_threshold that actually worked.

Adding your microphone is a genuinely useful two-minute contribution, and you do not need to know any Python. See Add yours below.

No entries yet — be the first

This page was created so people can append to it without waiting for anyone. If the table below is still empty when you read this, add the first row.

The list

Sorted by microphone name. Entries are contributors' own reports, not endorsements, and nothing here is tested by the maintainers.

Microphone Type OS / desktop vad_threshold Notes
(your entry here)

Add yours

  1. Measure your mic — this prints the number the table wants:

    yazses mic-level

    It records for a few seconds, reports your average level against the current threshold, and recommends a value. Add --set if you want it written to your config as well.

  2. Check your room while you're there — this warns if ambient noise is loud enough to trip the silence gate on its own:

    yazses doctor --mic

  3. Add one row to the table above, in alphabetical order by microphone name, and delete the _(your entry here)_ placeholder row if it is still present.

  4. Open a pull request titled docs: add <microphone> to known-good microphones.

What to put in each column

Column What to write
Microphone The model as it is sold — Blue Yeti, Jabra Evolve2 65, ThinkPad T14 built-in.
Type USB, built-in, XLR + interface, Bluetooth, or headset.
OS / desktop e.g. Ubuntu 26.04 / GNOME (Wayland), macOS 15, Windows 11. Wayland vs X11 matters.
vad_threshold What yazses mic-level recommended, or the value you settled on.
Notes The useful part — how far from your mouth, whether it needed the gain turned down, whether Bluetooth mode hurt quality.

Negative results are welcome and wanted. A microphone that transcribed badly, needed an odd threshold, or dropped to 8 kHz in headset mode is more useful to the next reader than another good result. Say so plainly in the notes.

Good to know

  • vad_threshold is machine-specific. It is compared against mean(|audio|) for your input, so a value that works on a quiet desktop may cut off speech in a noisy room. Treat the numbers here as starting points, not settings to copy blindly.
  • If dictation logs Silent audio -- discarding, your speech is falling below the gate — lower the threshold. If room noise produces spurious transcripts, raise it. See Troubleshooting.
  • YazSes can also pin an input so a newly-connected device cannot steal capture: yazses audio use "<name>". Worth noting in your entry if you needed it.