Live transcript previews Automatic (SpeechRecognition when supported) Browser SpeechRecognition Whisper (multilingual, on this device) Sherpa (English / Chinese, on this device) Off

Italic text is a temporary preview. The server supplies the final transcript. Whisper and Sherpa download models from this server. Browser SpeechRecognition may send audio to the browser’s speech service and uses the first language in a pair.

Catchup speed

1.5×

With WSOLA enabled, live catchup starts above 100 ms of buffered audio and uses the selected speed without raising pitch. It switches straight back to 1× when the buffer drains. Message replay catches up when more than three seconds behind. Set to 1× to disable acceleration.

WSOLA live catchup

Preserves voice pitch while catching up with live audio. Disable to keep live audio at 1×. Replay keeps the browser’s pitch preservation.

Volume threshold

0.018

With Silero off, microphone volume detects speech. Raise the threshold above background noise, or lower it for quiet speech. Changes apply immediately.

Silero speech detection

Enable in noisy environments to distinguish speech from other sounds instead of using volume. Adds processing delay. Does not identify a particular speaker or detect sentence completion.

Microphone gain

0 dB

Automatically adjust speech to −19 LUFS

Adjust outgoing voice volume. Automatic adjustment requires Silero and replaces manual gain while enabled. It learns from speech, holds during pauses, and settles gradually. Peak protection may keep loud passages below the target. Changes apply immediately.

Automatic gain: waiting for speech · Target −19 LUFS.

Speech confidence

0.10

Lower accepts quieter or uncertain speech; higher rejects more sounds but may miss speech.

Silence before closing speech gate

200 ms

Keeps the end of words during short pauses. With Silero on, samples finish after one second of silence. At nine seconds, longer samples split at their longest pause.

Browser noise suppression

Ask the browser to reduce microphone noise before advanced processing. Can be combined with either algorithm below; support and results vary by browser.

Advanced noise suppression No Advanced Noise Suppression RNNoise FastEnhancer FastEnhancer model Tiny — lowest CPU use Base — balanced Small — highest quality, more CPU

Apply RNNoise or FastEnhancer after browser processing to clean outgoing voice and transcription audio, or select no advanced suppression. Echo cancellation stays on. Changes briefly restart an active microphone.

FastEnhancer runs on your device. Choose Tiny for slower devices, Base for a balance of quality and speed, or Small for the strongest cleanup. Changing models briefly restarts an active microphone.