Skip to main content

Streaming Transcription service usage

Streaming speech to text standalone worker. Processes audio streams on an WS endpoint

WS to server​

Establish connection​

Initiate the server connection by sending a json message. {"config": {"sample_rate":16000, "metadata":1 }}

Option :

  • sample_rate
  • metadata :
    • 1 : application/json
    • 0 : text/plain

Sending audio​

After the connection is established, you can send an audio stream to the server.

Tested audio format :

  • format : 16-bit binary string
  • channels : 1
  • rate : 16000 (make sure the rate is the same as the one used initially)
  • buffer_size : 1024 (can be any integer)

Python example audio.open(format=pyaudio.paInt16, channels=1, rate=16000, input=True, frames_per_buffer=1024)

Ending session​

Send a {"eof" : 1} message to end the session.

Server to WS​

Response to an audio stream​

{"partial": "one"}
{"partial": "one sentence"}
{"partial": "one sentence of"}
{"partial": "one sentence of user"}

Response to an audio stream after a silence duration (~0.5s)​

{"text": "one sentence of user"}

Response after ending session​

The server will send after an {"eof" : 1} (depending of the initial metadata)

  • metadata 0 : text/plain
  • metadata 1 : {"speakers": [], "text": "", "words": []}

Finally, the server will close the connection