The Utterance End feature can be used to detect the end of speech by waiting a configured amount of milliseconds of silence after the last detected speech. The end of speech is detected using a combination between the VAD (Voice Activity Detection) and the transcription model. The transcription model is also used to prevent false positives of speech detection by VAD.

​
Configuration

The Utterance End feature can be configured by using the utteranceEnd=1000 query parameter, where 1000 is the amount of milliseconds of silence to wait after the last detected speech.

​
Result

When an utterance is ended, a transcription response with the "utterance": true attribute is emitted. The content of the response can be a transcription response or an empty transcription, depending on the underlying engine implementation. 
{
  "transcription": "",
  "words": [],
  "start": 6480,
  "end": 6480,
  "metadata": {},
  "utterance": true
}