Understanding Transcript Timing in Voice Conversations

When working with voice conversations that include both user and AI agent speech, you may need to track the timing of each segment in the transcript. Here's what you need to know about transcript timing and available options.

Current Functionality

When using the transcription_node with non-streaming TTS providers (like OpenAI TTS), the system currently:

• Provides start times for each sentence

• Does not provide explicit end times (chunk.end_time will show as NOT_GIVEN)

• The start time of the next sentence can be used as the end time of the previous sentence

Getting Complete Transcripts

To get a complete transcript with both user and AI speech:

• AI agent transcripts can be obtained from the transcription_node

• User transcripts are available through the user_input_transcribed event

Example Desired Output Format

{
  "transcript": [
    {
      "start_time": "00:00",
      "end_time": "00:04",
      "text": "Hi. Thanks so much for joining me today.",
      "role": "AI"
    },
    {
      "start_time": "00:11",
      "end_time": "00:15",
      "text": "I know me look online and online gift recommendations.",
      "role": "User"
    }
  ]
}