Speech recognition HTTP-server on NET7 using vosk/kaldi

Presets

Docker Desktop -> settings -> resources -> Advanced

change RAM to 8Gb+

Vosk/kaldi server

Use WebSocket server docker run -d -p 2700:2700 alphacep/kaldi-ru:latest. In project this server run from docker-compose

• https://alphacephei.com/vosk/install
• https://alphacephei.com/vosk/models
• https://github.com/alphacep/vosk-server
• https://github.com/alphacep/vosk-api

IMPORTANT: the server only processes audio in the format WAV, codec PCM, channels MONO 16bit, frequency 8khz

How it works

Run from directory of solution docker-compose up -d

Attach audio file and execute request:

curl -X 'POST' \
	'https://localhost/api/Speech' \
	-H 'accept: */*' \
	-H 'Content-Type: multipart/form-data' \
	-F 'file=@voxworker-voice-file_8khz.wav;type=audio/wav'

Response

{
  "items": [
    {
      "conf": 1,
      "end": 0.57,
      "start": 0,
      "word": "создать"
    },
    {
      "conf": 0.427799,
      "end": 1.062693,
      "start": 0.57,
      "word": "задачу"
    },
    {
      "conf": 1,
      "end": 1.47,
      "start": 1.08,
      "word": "купить"
    },
    {
      "conf": 1,
      "end": 1.89,
      "start": 1.47,
      "word": "творог"
    }
  ],
  "text": "создать задачу купить творог",
  "partial": null
}

Helpers

https://convertio.co/ - help convert files to the desired format