A very simple server based on Vosk-API.

There are three implementations for different protocol - websocket, grpc, mqtt.

Usage

Start the server

docker run -d -p 2700:2700 alphacep/kaldi-en:latest

or for Chinese. The model is based on Kaldi multi-cn recipe, thanks to Xingyu Na.

docker run -d -p 2700:2700 alphacep/kaldi-cn:latest

or for Russian

docker run -d -p 2700:2700 alphacep/kaldi-ru:latest

or for German (model from https://github.com/uhh-lt/kaldi-tuda-de#pretrained-models)

docker run -d -p 2700:2700 alphacep/kaldi-de:latest

Run

git clone https://github.com/alphacep/vosk-server
cd vosk-server/websocket
./test.py test.wav

You can try with any wav file which has proper format - 8khz 16bit mono PCM. Other formats has to be converted before decoding.

Troubleshooting

Server is compiled to work on modern CPU with AVX2 support for best decoding speed. If you want to use older CPU please try to use atom version like kaldi-en-atom instead of kaldi-en.

Some servers use huge carpa models for best accuracy. To load them in memory you need about 8Gb of memory or even more. Make sure you have enough memory on your server.

In case of problems try to run server manually from docker prompt and see what happens:

$ sudo docker run -it -p 2700:2700 alphacep/kaldi-en:latest /bin/bash
root@a9e0db45a54b:/opt/vosk-server/websocket# python3 ./asr_server.py /opt/vosk-model-en/model
LOG ([5.5.643~1-7e185]:ConfigureV2():src/model.cc:138) Decoding params beam=13 max-active=7000 lattice-beam=6
...

Testing with microphone

You would need to install the pyaudio pip package:

pip install pyaudio

(on Windows, it's easiest to install the wheel from here).

To test with a microphone, run

./test_microphone.py localhost:2700

Connected to ws://localhost:2700
Type Ctrl-C to exit
{"partial" : ""}
{"partial" : "поднимите мне веки"}
{"partial" : "поднимите мне веки"}
{"result" : [ {"word": "поднимите", "start" : 3.45, "end" : 4.26, "conf" : 1},
{"word": "мне", "start" : 4.26, "end" : 4.47, "conf" : 1},
{"word": "веки", "start" : 4.47, "end" : 5.07, "conf" : 1}
 ], "text" : "поднимите мне веки" }
{"partial" : ""}
Closing PyAudio Stream
Terminating PyAudio object
Terminating connection
{"result" : [  ], "text" : "" }
Bye

Other programming languages

Check other examples (Asterisk-EAGI, php, java, node, c#) in client-samples folder in this repository.