Websocket interface

The websocket interface enables you to add speech recognition capabilities to your client applications. The transcription service relies on websocket connection to asynchronously exchange bi-directional data (audio chunks and JSON text messages) in real-time.

API calls require a api key and language model specified as URL parameter.

Base URL for API calls is wss://developer.speech-i.com/ws/client/speech?key=XXX&model=YYY

 Usage

The recognition request and response cycle has the following steps:

1. The cient establishes a websocket connection to the server


2. On connection opened, client must send the JSON formatted transcription parameters


3. Server responds with a JSON formatted status message.
{"status":"ready|error|busy"}
4. If the server status is ready, the client can start sending the audio chunks in binary format (binary frame); to limit the latency, each chunck should contain at most 500ms of audio.
Audio format must be mono, encoding speex or raw PCM, sampling rate 8kHz or 16kHz. Other formats may be available on request.
5. Server periodically sends JSON formatted partial results, according to transcription parameters settings.
{"final":false,"segment-start":0,"total-length":1.75,"transcript":"hello"} {"final":false,"segment-start":0,"total-length":2.75,"transcript":"hello world"}
6. On audio transfer completed, the client can send a stop decoding message.
{"decoding":"stop"}
7. Server sends the JSON formatted final transcription.

{
  "confidence":0.967202366309009,
  "final":true,
  "hypotheses":[
    {
      "likelihood":135.98733520507813,
      "transcript":"hello world",
      "word-alignment":[
        {
          "end":0.3899999912828207,
          "start":0,
          "word":"hello"
        },
        {
          "end":1.409999968484044,
          "start":0.41999999061226847,
          "word":"world"
        }
      ]
    },
    {
      "likelihood":133.66854858398438,
      "transcript":"yellow world"
    }
  ],
  "segment-length":1.409999968484044,
  "segment-start":0
}

8. According to transcription parameters, the iteration can restart from point 3.

 Transcription parameters

The following table shows the customizable parameters that client must send when websocket connection is opened.

 Language models

The following table shows supported language models (other languages are available on request):