Home Assistant changes how it converts streaming speech for voice satellites, letting replies begin without waiting for most of the message to arrive. The change applies to streaming speech from Wyoming and Home Assistant Cloud when the audio needs conversion.
In Michael Hansen’s simulated three-second utterance, generated at just above speaking speed, the first converted audio arrived after 0.10 seconds instead of 2.90. These are the best of three runs at the conversion stage, rather than the full time between a spoken request and its answer.
Hansen reads the incoming audio format first and passes its samples directly to the converter, avoiding the wait for it to inspect the stream. Complete audio files retain their existing handling.