curl --request POST \
  --url http://{device-ip}:{device-port}/audio-analytics/v1/api/transcriptions/create \
  --header 'Content-Type: multipart/form-data' \
  --form model=whisper-small-quantized \
  --form file='@example-file' \
  --form stream=false \
  --form language=en \
  --form 'parameters[0]={
  "key": "sampling_rate",
  "value": "16000"
}' \
  --form 'parameters[1]={
  "key": "channels",
  "value": "1"
}'

{
  "text": "The transcribed text from the audio file.",
  "language": "en",
  "type": "transcript.text.done"
}

Transcribe audio to text

Transcribes audio input into text. Supports three modes:

1. File Upload (Non-Streaming)

Upload an audio file and receive the complete transcription in a single response.

Request:

Content-Type: multipart/form-data
Include file, model, stream=false

Response:

Complete transcription text
Detected language

2. File Upload (Streaming Results)

Upload an audio file and receive partial transcription results via WebSocket.

Request:

Content-Type: multipart/form-data
Include file, model, stream=true

Response:

Initial acknowledgment with session_id
Partial results via WebSocket (transcript.text.delta)
Final result via WebSocket (transcript.text.done)

3. Live Audio Streaming

Stream audio chunks in real-time via WebSocket.

Request:

Content-Type: application/json
Include model, stream=true (no file)

Response:

Session ID for WebSocket connection
Send audio chunks via WebSocket
Receive events and partial results in real-time

Event Sequence (Live Streaming with VAD)

When streaming live audio with VAD/EPD enabled, the server emits the following sequence of WebSocket messages for each utterance:

transcript.event / speech_start — speech first detected by VAD

{ "session_id": "session123", "type": "transcript.event", "state": "speech_start" }

transcript.text.delta — zero or more partial transcription results

{ "session_id": "session123", "text": "Partial text...", "language": "en", "type": "transcript.text.delta", "state": "transcription" }

transcript.event / speech_end — end-of-speech detected by VAD/EPD

{ "session_id": "session123", "type": "transcript.event", "state": "speech_end" }

transcript.text.done — final transcription for the utterance

{ "session_id": "session123", "text": "Complete utterance text.", "language": "en", "type": "transcript.text.done", "state": "transcription" }

After step 4 the engine resets and listens for the next utterance. The session remains open until the client calls /transcriptions/close. The client may also call /transcriptions/flush at any time to force immediate processing of buffered audio.

Important - Session Timeout: When streaming audio, if no audio data is sent for a period of time (session timeout), the session will automatically close and the final transcription will be sent. This prevents sessions from remaining open indefinitely when audio streaming has stopped.

Supported Audio Formats

WAV (recommended)

Sample Rates

16000 Hz (recommended)
Automatic resampling available

Parameters

You can include additional parameters for fine-tuning:

sampling_rate: Target sample rate (e.g., “16000”)
channels: Number of audio channels (e.g., “1” for mono)

POST

transcriptions

create

curl --request POST \
  --url http://{device-ip}:{device-port}/audio-analytics/v1/api/transcriptions/create \
  --header 'Content-Type: multipart/form-data' \
  --form model=whisper-small-quantized \
  --form file='@example-file' \
  --form stream=false \
  --form language=en \
  --form 'parameters[0]={
  "key": "sampling_rate",
  "value": "16000"
}' \
  --form 'parameters[1]={
  "key": "channels",
  "value": "1"
}'

{
  "text": "The transcribed text from the audio file.",
  "language": "en",
  "type": "transcript.text.done"
}

Body

model

string

required

ID of the ASR model to use

Example:

"whisper-small-quantized"

file

Audio file to transcribe (for file-based transcription)

language

string

ISO 639-1 language code (e.g., 'en' for English). If omitted, language may be auto-detected.

Example:

"en"

stream

default:false

If true, results will be streamed via WebSocket. If false, complete transcription returned in response.

Example:

false

channels

Number of audio channels in the input stream (1=mono, 2=stereo). Takes priority over the channels key inside the parameters array.

Example:

1

parameters

Additional parameters for transcription (e.g., sampling_rate, channels)

Example:

"[{\"key\": \"sampling_rate\", \"value\": \"16000\"}, {\"key\": \"channels\", \"value\": \"1\"}, {\"key\": \"server_timeout\", \"value\": \"60\"}]"

Response

Transcription request accepted

Option 1
Option 2

Transcription result (non-streaming)

text

string

required

The transcribed text

Example:

"The sky is blue and the sun is shining."

language

string

Detected or specified language (ISO 639-1 code)

Example:

"en"

language_name

string

Detected language full name as returned by the ASR engine

Example:

"English"

type

enum<string>

Result type

Available options:

transcript.text.done,

transcript.text.delta

Example:

"transcript.text.done"

session_id

string

Session identifier (if applicable)

Example:

"session123"

List available ASR models Close active transcription session