WebSocket protocol

Authentication, client messages, server events, and audio requirements for realtime transcription.

Connect to:

wss://api.orbitalsai.com/api/v1/streaming/transcribe?token=<API_KEY>

Send PCM16 little-endian, mono audio at 16 kHz as binary frames. Send control messages as JSON text frames.

Client messages

Configure the session after receiving ready:

{
  "type": "config",
  "language": "hausa",
  "sample_rate": 16000,
  "return_timestamps": true,
  "translate": false
}

To enable translation, set translate to true and provide target_language. Optional fields include translation_domain and translation_window_ms.

When the audio source ends, send {"type":"flush"} and wait for flushed before closing.

Server events

Event	Meaning
`ready`	The connection is authenticated and ready.
`speech_start` / `speech_end`	Voice activity state changed.
`partial`	Replaceable transcript draft.
`final`	Stable transcript segment; may include word timestamps when requested.
`translation`	Translation for one or more finalized segment IDs.
`translation_failed`	Translation failed; the source final remains valid.
`flushed`	Buffered audio has been finalized.
`error`	The session cannot process a request or continue.

See Realtime transcription, Realtime translation, and the Python realtime SDK.

Client messages

Server events

On this page