Automatically end the session if nobody speaks for this long. Useful when you forget to tap Stop.

Larger models are more accurate but slower and use more memory. Applies on the next session start.

How long of a pause ends an utterance. Shorter = chunks more often; longer = fewer fragments mid-sentence.

Minimum loudness counted as speech. Raise this in noisy environments to avoid false triggers.

Audio captured before each speech onset. Increase if first words get clipped.