Installing Kaldi

[gaurav gund]

Hi Povey! 

I am able to use pretrained kaldi models. I am planning to do training but installing Kaldi is so time-consuming and doing it on google colab is painful. Are there any workaround to reduce the time for kaldi installation? Are there any ways to avoid installing it everytime a google colab sessions is started.

Also, any full-fledge DNN based tutorial from scratch with code snippet will be helpful! 

Regards

Gaurav

[Daniel Povey]

Perhaps colab is not the best place to run Kaldi, some kind of linux machine, in the cloud or not, might be more suitable.

the mini_librispeech recipe is pretty easy to run from scratch- but any DNN training will require access to a GPU.

Dan

--
Go to http://kaldi-asr.org/forums.html to find out how to join the kaldi-help group
---
You received this message because you are subscribed to the Google Groups "kaldi-help" group.
To unsubscribe from this group and stop receiving emails from it, send an email to kaldi-help+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/kaldi-help/CALU0U8mXrbLrv4aEZYLiQgOYJ1%3Dn3aa3K%2BzWAPCNSsxxd3aKrg%40mail.gmail.com.

[gaurav gund]

Will it be possible to train DNN using free GPU on colab?

Regards

To view this discussion on the web visit https://groups.google.com/d/msgid/kaldi-help/CAEWAuyT8nz-f46OzqrUoTXfzmW6svdcnLEpGKQemORK6OrTwxQ%40mail.gmail.com.

[itcor...@gmail.com]

It seems to be a little difficult to train on free Google Colab.

I recommend you rent such a GPU machine on https://vast.ai.

Best,

Charles.

[gaurav gund]

Hi charles,

Could you suggest some tutorial for simpler models, I have aleeady tried pretrained models for decoding.

Regards

To view this discussion on the web visit https://groups.google.com/d/msgid/kaldi-help/cade9394-7a1b-40e1-b336-d1e6f332df31n%40googlegroups.com.

[gaurav gund]

Hi,

Is there a way to speed-up “!make” commands while install kaldi?

I already tried ‘! make -j 4’ but all in vain.

Regards

Gaurav

[Daniel Povey]

probably not, other than decreasing optimization level, e.g. change -O1 to -O0 in kaldi.mk

or use larger -j option

To view this discussion on the web visit https://groups.google.com/d/msgid/kaldi-help/CALU0U8mVm4_ZZr%3Das0shFSoFfukfm6Y4WA%2BypQ92E_0m3ETQkA%40mail.gmail.com.