AldeaConceptsStreaming controls
Endpointing
Configure speech endpoint detection with endpointing parameter
Endpointing controls how quickly the API finalizes a sentence during streaming. It determines how long the server waits after the last detected speech before marking a result as is_final: true.
Usage
Set the endpointing parameter in the WebSocket query string. The value is in milliseconds:
# Faster finalization (300ms)
wss://api.aldea.ai/v1/listen?endpointing=300&encoding=mp3
# Slower finalization (800ms), better for speakers who pause mid-sentence
wss://api.aldea.ai/v1/listen?endpointing=800&encoding=mp3To disable endpointing entirely, set it to false:
wss://api.aldea.ai/v1/listen?endpointing=false&encoding=mp3When endpointing is disabled, results are only finalized when you send a Finalize or CloseStream message.
Choosing a value
| Value | Use case |
|---|---|
100–300 ms | Live captions, real-time chat. Prioritize speed |
300–500 ms | General purpose. Balanced speed and accuracy |
500–1000 ms | Speakers who pause often. Avoid premature sentence breaks |
false | Manual control. Finalize only on explicit command |
Lower values produce faster results but may split sentences mid-thought. Higher values wait longer, producing more complete sentences at the cost of latency.