ESPHome fixes a failure in which a voice assistant heard a request but never spoke its answer. With a fast speech-recognition engine, response events could arrive before the microphone had actually stopped, leaving it in control of the audio connection the speaker needed.

The device waits for the microphone to stop before starting its reply. The correction was tested on an ESP32-S3-BOX-3 with its ready-made voice-assistant configuration and addresses the failure seen with Home Assistant Link’s second-generation speech recognition.