The Streaming STT API uses WebSockets for real-time transcription. You connect once, stream audio chunks, receive partial transcripts as they become available, and send a commit message to receive the final transcript.

​
Endpoint

wss://infer.voice.intron.io/stt/v1/stream

​
Authentication

All connections require a Bearer token in the Authorization header: 
Authorization: Bearer YOUR_API_KEY

​
Connection query parameters

The websocket route reads these values from request.args.get(...).
ParameterTypeDescriptionRequiredDefault
session_idStringExisting session ID. If omitted, server creates one and returns it in SESSION_CREATED.NoAuto-generated
sample_rateIntegerInput audio sample rate in Hz.No16000
bit_rateIntegerInput PCM bit depth.No16
num_channelsIntegerNumber of input channels.No1
use_language_asr_inputStringInput language code.(see the supported languages)Noen

​
Supported Languages

Supported Languages

Comprehensive list of languages supported

​
WebSocket handshake examples

wscat -c 'wss://infer.voice.intron.io/stt/v1/stream?sample_rate=16000&bit_rate=16&num_channels=1&use_language_asr_input=en' \
	-H 'Authorization: Bearer YOUR_API_KEY'

​
Input message types

​
INPUT_AUDIO_CHUNK

FieldTypeRequiredDescription
message_typeStringYesMust be INPUT_AUDIO_CHUNK
audio_base_64StringYesBase64 encoded PCM16 little-endian audio bytes
ack_idIntegerNoClient chunk sequence ID echoed by server ACK
JSON
{
	"message_type": "INPUT_AUDIO_CHUNK",
	"audio_base_64": "<base64_pcm16_chunk>",
	"ack_id": 1
}

​
COMMIT

FieldTypeRequiredDescription
message_typeStringYesMust be COMMIT
JSON
{
	"message_type": "COMMIT"
}

​
Response message types

Output payloads are based on websocket response classes (WebsocketResponse... and stream response classes).

​
Session and transcript messages

Sent immediately after successful authentication, capacity checks, and quota checks. It returns the active session_id, current credit balance, and the effective stream configuration.
JSON
{
	"message_type": "SESSION_CREATED",
	"session_id": "12a9760f-b165-4404-91d0-a65d4cdt78fs",
	"credit_balance": 120.0,
	"configs": {
		"sample_rate": 16000,
		"bit_rate": 16,
		"num_channels": 1,
		"use_prompt_id": null,
		"use_language_asr_input": "en",
		"use_language_asr_output": "en"
	}
}
Sent after the server accepts and stores an audio chunk. The server echoes the client ack_id when present, otherwise it uses the running chunk count.
JSON
{
	"message_type": "AUDIO_CHUCK_ACK",
	"chunck_id": 1,
	"total_chunks": 1
}
Sent whenever a new partial transcript is available and differs from the last partial transcript already sent on the connection.
JSON
{
	"message_type": "PARTIAL_TRANSCRIPT",
	"transcript": "patient reports intermittent chest pain"
}
Sent after the client sends COMMIT and the backend finishes processing the full stream. This is the final transcript payload for the session.
JSON
{
	"message_type": "COMMITTED_TRANSCRIPT",
	"transcript_id": "e1c4a90b-e319-4de6-9f22-0e0cf5e8b7a2",
	"transcript_text": "patient reports intermittent chest pain for two days",
	"audio_len": 10
}
Optional keep-alive message sent by the server when periodic pinging is enabled for the connection.
JSON
{
	"message_type": "PING",
	"message": "ping"
}

​
Error and terminal messages

Sent when the Authorization header is missing, invalid, or the access key fails authentication. The server closes the connection after sending this message.
JSON
{
	"message_type": "AUTHENTICATION_ERROR",
	"message": "permission denied,access-key error"
}
Sent when the service does not currently have enough available capacity to accept the stream. The status field indicates the capacity state returned by the backend.
JSON
{
	"message_type": "RESOURCE_EXHAUSTED",
	"status": "CAPACITY_NOT_AVAILABLE"
}
Sent when the authenticated user or entity has insufficient credits or has exceeded an allowed usage limit. The server closes the connection after sending this message.
JSON
{
	"message_type": "QUOTA_EXCEEDED",
	"credits_balance": "0",
	"message": "insufficient credits balance"
}
Sent when the request payload is invalid, such as malformed JSON, an invalid message structure, invalid base64 audio data, invalid PCM16 payload length, or audio sent after COMMIT.
JSON
{
	"message_type": "INPUT_ERROR",
	"message": "Invalid base64 audio payload"
}
Sent when the received audio chunk is smaller than the minimum allowed chunk size.
JSON
{
	"message_type": "CHUNCK_SIZE_TOO_SMALL",
	"chunk_size_min": "1"
}
Sent when the received audio chunk is larger than the maximum allowed chunk size.
JSON
{
	"message_type": "CHUNK_SIZE_TOO_LARGE",
	"chunk_size_max": "32"
}
Sent when no audio chunk has been received for longer than the allowed idle timeout while the session is still active.
JSON
{
	"message_type": "INSUFFICIENT_AUDIO_ACTIVITY",
	"message": "insufficient audio activity detected,exceed max idle time of 60 seconds"
}
Sent when the websocket session exceeds the maximum allowed session lifetime.
JSON
{
	"message_type": "SESSION_TIME_LIMIT_EXCEEDED",
	"session_time_limit": "300"
}
Sent when the server detects that the websocket connection has been lost.
JSON
{
	"message_type": "CONNECTION_LOST",
	"message": "connection lost"
}
Sent for unexpected server-side failures that do not map to a more specific websocket error message.
JSON
{
	"message_type": "ERROR",
	"message": "connection error"
}

​
Runtime limits

Max session lifetime: 300 seconds
Max idle audio gap: 60 seconds
Min chunk size: 1 KB
Max chunk size: 32 KB

​
Typical flow

  1. Open websocket connection with query parameters and Authorization header.
  2. Receive SESSION_CREATED.
  3. Send INPUT_AUDIO_CHUNK messages repeatedly.
  4. Receive AUDIO_CHUCK_ACK and PARTIAL_TRANSCRIPT messages.
  5. Send COMMIT when done streaming audio.
  6. Receive COMMITTED_TRANSCRIPT and the connection closes.