Mismatch features: 40 expected but 13 given

[chaoticz044]

Hi Devs,
I made an acoustic and language model using the scripts from steps and utils folder copied from wsj and used the nnet2 to train pnorm. I successfully made completed the training and decoding. I tried moving it to kaldi gstreamer but when I run the streaming decode result. I got logs from workers that feature dimensions are not the same. Expected is 40 but got 13. I used the same config in training locally. Do I need to convert the nnet2 model to online? If yes, how would I convert it and run it on gstreamer since I matched the configs in the yaml file and set the config files to the yaml file.

This is my yaml file:

use-nnet2: True
decoder:
    use-threaded-decoder: True
    nnet-mode: 2
    model : /opt/models/final.mdl
    word-syms : /opt/models/fst/words.txt
    fst : /opt/models/fst/HCLG.fst
    mfcc-config : /opt/models/mfcc.conf
    max-active: 7000
    beam: 25.0
    lattice-beam: 13.0
    acoustic-scale: 0.1
    do-endpointing : true
    endpoint.silence-phones: 1:2:3:4:5:6:7:8:9:10
    traceback-period-in-secs: 1.5
    chunk-length-in-secs: 0.1
    num-nbest: 3

    word-boundary-file: /opt/models/fst/word_boundary.int

out-dir: tmp

and my config file is:

--use-energy=false
--num-ceps=13

I already used the

steps/online/nnet2/prepare_online_decoding.sh

but not solved.

Please guide me on how to solve it. I used 13 mfcc and 3 steps/train_deltas.sh and lda.

Thanks a lot!

[svenha]

Your mfcc.conf might be the problem. Do you have other files matching the pattern `mfcc*conf?