search

alphacep / vosk-server

WebSocket, gRPC and WebRTC speech recognition server based on Vosk and Kaldi libraries
Apache License 2.0
918 stars 248 forks source link

model size and location #115

Closed MikeyBeez closed 3 years ago

MikeyBeez commented 3 years ago

Which model is being used? Can I use a larger model? How would I do that? Cheers. Thanks.

MikeyBeez commented 3 years ago

I'd like to use the best/largest model for the most accurate results. How do I do that?

nshmyrev commented 3 years ago

You download the model and unpack in the same folder as the server. Then run ./asr_server.py

MikeyBeez commented 3 years ago

So I found the model by looking in the container. It's the Aspire model, which appears to be your best model. Thanks.

nshmyrev commented 3 years ago

It's the Aspire model, which appears to be your best model.

aspire model is not the best

MikeyBeez commented 3 years ago

./asr_server.py

I think that means go into the container with docker exec -it . then cd /opt. Then download and unzip the model. It should overwrite vosk-model-en. But then it will not persist, unless I mount that as a volume. Otherwise, I could do all this in my dockerfile. Sorry for being so slow about this. I'm new to Docker.

MikeyBeez commented 3 years ago

If aspire isn't the best, which is? These are my choices: https://alphacephei.com/vosk/models

MikeyBeez commented 3 years ago

I found these too: http://kaldi-asr.org/models/m13

nshmyrev commented 3 years ago

But then it will not persist, unless I mount that as a volume.

You can simply mount your model as a volume instead of /opt/vosk-model-en.

It is also easy to create your own container which inherits kaldi-en.

If aspire isn't the best, which is?

There are numbers in the table.