Automatic Speech Recognition Sample example using "Kaldi decoder » : Nnet-forward Kaldi doesn’t work

[MGabi]

Hi,

I'm on ubuntu 18.04, I work with Openvino 2020.1.023, Kaldi version 5.5, Python* version 3.6.9 and Cmake* version 3.10.2.

I would like to use the Speech_SampIe function, for this I need an ark file as input. So I tried to execute the following command line: nnet-forward --use-gpu=no final.feature_transform "ark,s,cs:copy-feats scp:wav.scp ark:- |" ark:feat.ark

I found this one in the Use of Sample in Kaldi* Speech Recognition Pipeline section on the following web site :

https://docs.openvinotoolkit.org/2018_R5/_samples_speech_sample_README.html

nnet-forward terminate called after throwing an instance of 'kaldi::KaldiFatalError' . Find attached the error message.

I think that i put the wrong scp file but i don’t know how to create the right scp file from wav file. Have you got some documentation to do it ?

Best regards,

Martial

[David_C_Intel]

Hi MGabi,

Thanks for reaching out.

This seems like an issue in using the Kaldi s5 recipe. The support for this is beyond the scope of OpenVINO™ toolkit support. Kaldi related communities might help you, such as this github forum.

Best regards,

David C.