search

kaldi-asr / kaldi

kaldi-asr/kaldi is the official location of the Kaldi project.
http://kaldi-asr.org
Other
14.03k stars 5.3k forks source link

If I use the pretrained model but the input-dim isn't match ,how can i adjust input-dim of my wavs? #4843

Open baowenqian2001 opened 1 year ago

baowenqian2001 commented 1 year ago

ERROR (nnet3-compute[5.5.1061~2-e4eb]:EnsureFrameIsComputed():nnet-am-decodable-simple.cc:101) Neural net expects 'input' features with dimension 43 but you provided 40

desh2608 commented 1 year ago

You should use the same feature config file (conf/mfcc.conf) to extract features for your wav, as the one which was used to train the model. In this case, it seems the model was trained using hires+pitch features, but you are just providing it hires features. You should use make_mfcc_pitch.sh to extract features instead of make_mfcc.sh. You would also need to add a pitch.conf in your conf directory (search for this file in kaldi/egs for an example).

baowenqian2001 commented 1 year ago

Appreciate it so much. It's useful!