Improvements
- Adjustable Pre-Speech Padding for the Free provider: a new slider keeps up to 2 s of audio from just before speech is detected, so the first sound of a sentence is not cut off; the default stays at 0.8 s. It is shown for the models that support it (#603).
- Timed diagnostic logs: every line copied from the diagnostic logs now carries a time in milliseconds, so a log attached to a bug report shows where a delay comes from (#605).
Fixes
- The Free provider's speech detection missed speech and let background noise through: its detector was fed audio without the context its model expects. It now catches more of what you say and passes far less noise to recognition (#601).
Full Changelog: v0.43.1...v0.43.2