vosk-server
vosk-server copied to clipboard
With three clients running at the same time, the kaldi-en-gpu container stops.
docker run --gpus=all -p 2700:2700 alphacep/kaldi-en-gpu
Running only one client, it completed successfully after 4 minutes.
cd /root/vosk-server/websocket && time ./test.py test2.wav
Running three clients at the same time, an error is reported after 1 minute.
cd /root/vosk-server/websocket && time ./test.py test2.wav
cd /root/vosk-server/websocket && time ./test.py test2.wav
cd /root/vosk-server/websocket && time ./test.py test2.wav
docker logs
INFO:root:Connection from ('172.17.0.1', 45382)
INFO:root:Config {'sample_rate': 44100}
INFO:root:Connection from ('172.17.0.1', 45406)
INFO:root:Config {'sample_rate': 44100}
INFO:root:Connection from ('172.17.0.1', 45410)
INFO:root:Config {'sample_rate': 44100}
INFO:root:Connection from ('172.17.0.1', 45414)
INFO:root:Config {'sample_rate': 44100}
WARNING ([5.5.1027~1-59386]:OutputArcForce():word-align-lattice.cc:577) Invalid word at end of lattice [partial lattice, forced out?]
WARNING ([5.5.1027~1-59386]:OutputSilenceArc():word-align-lattice.cc:366) Phone changed before final transition-id found [broken lattice or mismatched model or wrong --reorder option?]
WARNING ([5.5.1027~1-59386]:OutputSilenceArc():word-align-lattice.cc:366) Phone changed before final transition-id found [broken lattice or mismatched model or wrong --reorder option?]
Looking forward to someone finally fixing this.
the problem is related as all the audios use the same websocket endpoint and that is not possible, you need modify the path to each audio use his own path... when you send multiple audios into the same path, brake the time logic of the audio flow... and that is the reason of the stop.
examples script only be prepared to run 1 audio at a time... if you need multiple audios, need modify the example