Principle:Elevenlabs Elevenlabs python Transcript Event Handling
| Knowledge Sources | |
|---|---|
| Domains | Speech_Recognition, Event_Driven_Architecture |
| Last Updated | 2026-02-15 00:00 GMT |
Overview
An event-driven architecture for processing real-time transcription results through registered callback handlers, supporting partial transcripts, committed transcripts, error events, and session lifecycle events.
Description
Transcript Event Handling provides the mechanism for applications to react to real-time transcription results. The system uses an observer pattern where callbacks are registered for specific event types, and the WebSocket message handler dispatches incoming messages to matching callbacks.
The event system defines multiple event types:
- Session lifecycle: open, close, session_started
- Transcription results: partial_transcript, committed_transcript, committed_transcript_with_timestamps
- Error events: error, auth_error, quota_exceeded, rate_limited, transcriber_error, and more
- Flow control: commit_throttled, queue_overflow, resource_exhausted
Specific error events also emit a generic ERROR event, allowing a single error handler to catch all error types.
Usage
Use this principle when consuming results from a realtime STT connection. Register handlers for the events you care about using connection.on(event, callback). At minimum, register handlers for COMMITTED_TRANSCRIPT and ERROR events.
Theoretical Basis
The event handling follows the Observer (publish-subscribe) pattern:
# Abstract event system
connection.on("partial_transcript", lambda data: update_live_caption(data))
connection.on("committed_transcript", lambda data: save_final_text(data))
connection.on("error", lambda data: log_error(data))
# Internal: WebSocket message loop dispatches to handlers
for message in websocket:
event_type = message["message_type"]
for handler in handlers[event_type]:
handler(message)
Key design decisions:
- Multiple handlers per event type (append-only registration)
- Synchronous handler execution (handlers should be fast)
- Specific error events also emit generic ERROR for unified error handling