search

NVIDIA / NeMo

A scalable generative AI framework built for researchers and developers working on Large Language Models, Multimodal, and Speech AI (Automatic Speech Recognition and Text-to-Speech)
https://docs.nvidia.com/nemo-framework/user-guide/latest/overview.html
Apache License 2.0
11.96k stars 2.49k forks source link

How should i use Kaldi MFCCs? #4314

Closed mesut92 closed 2 years ago

mesut92 commented 2 years ago

Hi I have kaldi feats.ark and text, but i do not have wavs. And i need to train ASR. I guess you had scripts for that before, but you removed them. Do you have any suggestion for me?Or any way to use feats?

Thanks Best Regards

https://docs.nvidia.com/deeplearning/nemo/user-guide/docs/en/stable/starthere/best-practices.html?highlight=kaldi#using-kaldi-formatted-data

titu1994 commented 2 years ago

We don't support kaldi data readers in Nemo 1.x, since there was no demand for it over several years.

titu1994 commented 2 years ago

You could convert your data to Nemo compatible manifest format which would make it possible to train ASR models.

mesut92 commented 2 years ago

But manifest format requires wavs, wright ?

titu1994 commented 2 years ago

Hmm right.

We deprecated this around the beginning of Nemo 1.x and it's been a long time so I don't think we will add back kaldi support soon unless there is some significant number of people who request it.

We recently dropped another kaldi io library because it was not maintained in over 3-4 years.

mesut92 commented 2 years ago

Thanks for quick response.