nnet online decoding without ivectors

alumae / kaldi-gstreamer-server

Real-time full-duplex speech recognition server, based on the Kaldi toolkit and the GStreamer framwork.

BSD 2-Clause "Simplified" License

1.07k stars 341 forks source link

nnet online decoding without ivectors #153

Open Umar17 opened 5 years ago

Umar17 commented 5 years ago

Is it possible to use gst-kaldi-nnet2-online for a nnet model trained without extracting ivectors? It throws no error when I comment out ivector-extraction-config line in worker's configuration, but the decoded output is just junk. Is the problem is model without ivectors or something else?

Best Regards

alumae commented 5 years ago

It should be possible but I have never tried. Does the model work fine when decoding using Kaldi's native tools?

Umar17 commented 5 years ago

Yes. Decoding is fine with kaldi's decoding script. If I don't comment out ivector-extraction-config in configuration it raises: ERROR ([5.2]:Compute():nnet-compute-online.cc:55) Feature dimension is 113, but network expects 13