Skip to content
Gladia Help Center home
Gladia Help Center home

Handle Real-Time transcript events

The realtime transcript event is the core message you’ll consume during a live session - https://docs.gladia.io/api-reference/v2/live/message/transcript

Every time the model detects speech, it streams transcript updates containing:

  • utterance: includes language, start/end timestamps, word-level timings and confidences.

  • is_final: a boolean flag indicating whether the transcript is still partial (false) or finalized (true).

You will typically receive:

  • One or more partial messages (is_final: false) for the same utterance.

  • A final message (is_final: true) that confirms the full transcript for that utterance.

A partial message is a preliminary result from the Real-Time API, while the server is working on doing a better transcription. When is_final=true, it means the model has finalized the transcription of this utterance, and will start a new one if a new speech occurs.

When handling messages, use the data.id to identify and update utterances:

  • Render or replace partials as they arrive.

  • Trigger downstream processing only when the final message is received.

To enable partials transcript, please add this to your configuration :

"messages_config": {
   "receive_partial_transcripts": true,
}

To learn more about the partials : https://docs.gladia.io/chapters/live-stt/features#partial-transcripts

Custom utterance segmentation

Two parameters control how utterances are segmented:

  • endpointing: defines the silence duration (in seconds) that signals the end of speech.

    • The default value is 0.05, meaning that the utterance is finalized after ~50 ms of silence.

  • maximum_duration_without_endpointing: sets a hard limit (in seconds) for utterance length.

    • Default: 5 seconds.

    • Ensures an utterance is finalized even if no silence is detected.

Play with those parameters to find the more suitable values for your use case.

Configuration

You can further customize transcript delivery via session configuration:

  • messages_config.receive_partial_transcripts → enable/disable partial messages.

  • messages_config.receive_final_transcripts → enable/disable final messages.

Powered by Plain