Closed YunzhaoLu closed 6 years ago
This asr server is created with Nnet3 decoder (see Nnet3LatgenFasterDecoder.cc).
Most likely you will need to adopt the code to use the same decoder as in kaldi lre07. You will have to build the server using it instead of Nnet3LatgenFasterDecoder and ensure it properly initialized and used.
Also from looking into https://github.com/kaldi-asr/kaldi/blob/master/egs/lre07/v2/run.sh#L202, they seem to use additional lineir regression classification on top of kaldi output. You will have to incorporate it somehow too.
So pretty what ever is happening during this script https://github.com/kaldi-asr/kaldi/blob/master/egs/lre07/v2/local/lre07_eval/lre07_eval.sh need to be happening inside this server. Not an easy task.
Thank you for the advice.
Hi, sorry to bother with such a question. I used kaldi to train lre07 to get the models to identify spoken languages. I want to apply the models for online recognizing languages. And I think asr-server is good tool to extend to support online language identification with online ivector extraction. Would you please give me some guidance how to do this? Thank you very much.
Regards, Luke