Speech to Text (WebSocket)

Real-time speech-to-text over WebSocket. Send audio chunks as you record and receive interim transcripts as the speech is recognized.

Connection

wss://api.async.com/speech_to_text/stream?api_key=sk_your_token&model_id=async_asr_v1.0

Query Parameters:

Parameter	Required	Description
api_key	Yes	Your API key (`sk_...`)
model_id	No	Model to use. Default: `async_asr_v1.0`
language	No	Language hint (ISO 639-1 code)

💡 See Speech to Text for detailed parameter descriptions.

Audio Format

Send audio as PCM16, 16 kHz, mono — either:

Binary frames: raw PCM16 bytes (preferred for performance)

JSON frames: {"audio": "<base64-encoded PCM16 chunk>"}

Send chunks of approximately 0.25–0.5 seconds (4000–8000 bytes at 16 kHz PCM16).

Signaling End of Audio

When you're done sending audio, send a text frame:

{"final": true}

This tells the server to finalize the transcript and close the stream.

Server Messages

The server sends JSON text frames:

Interim result (partial transcript)

{
  "type": "interim",
  "text": "Hello how are you doing",
  "delta": "doing"
}

text — cumulative transcript so far (replace your UI text with this)

delta — only the new piece since the last message (append to your UI)

Use whichever field fits your UI pattern.

Final result

{
  "type": "done",
  "text": "Hello, how are you doing today?"
}

Sent after receiving {"final": true}. Contains the finalized, punctuated transcript.

Error

{
  "type": "error",
  "error": "Description of what went wrong"
}

Examples

Python

JavaScript (Browser — live microphone)

Notes

Send audio continuously as it's captured — don't wait for silence.

The server sends interim messages as it recognizes speech. These are cumulative (the text field always contains the full transcript so far).

After you send {"final": true}, the server responds with a done message containing the finalized transcript, then closes the connection.

Cold start: the first connection after idle may take 60–150s. Keep the connection open during a session to avoid repeated cold starts.

Speech to Text (WebSocket)

Connection#

Audio Format#

Signaling End of Audio#

Server Messages#

Interim result (partial transcript)#

Final result#

Error#

Examples#

Python#

JavaScript (Browser — live microphone)#

Notes#

Request