Open pallav11 opened 4 years ago
Hi,
You can refer to egs/rm setup which contains transfer learning scripts ( https://github.com/kaldi-asr/kaldi/blob/master/egs/rm/s5/local/chain/tuning/run_tdnn_wsj_rm_1a.sh ). Using this script you can train the source model on a large amount of data and tune it with the target Indian accent data.
Thanks, Ashish
On Mon, Oct 12, 2020 at 1:12 AM pallav11 notifications@github.com wrote:
Hi,
I have data on Indian english accent. Kindly suggest how i can use Kaldi using transfer learning.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/kaldi-asr/kaldi/issues/4292, or unsubscribe https://github.com/notifications/unsubscribe-auth/AHAUCRI5G3CFCXKHZ7VGIDLSKKF3DANCNFSM4SMJY24Q .
Hi Ashish,
Thanks for the guidance, do you have any other documentation for using the same script as mentioned.
As the current documentation is not very clear.
If you like, you can look at the run.sh https://github.com/aarora8/kaldi2/blob/opensat_oct2020/egs/OpenSAT2020/s5/run_fine_tune.sh script and finetune_script https://github.com/aarora8/kaldi2/blob/opensat_oct2020/egs/OpenSAT2020/s5/local/chain/run_finetune_tl.sh. I would be happy to help with it. After some testing and review, we will try to put it into kaldi/master soon. Thanks, Ashish
On Mon, Oct 26, 2020 at 2:26 AM pallav11 notifications@github.com wrote:
Hi Ashish,
Thanks for the guidance, do you have any other documentation for using the same script as mentioned.
As the current documentation is not very clear.
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/kaldi-asr/kaldi/issues/4292#issuecomment-716333769, or unsubscribe https://github.com/notifications/unsubscribe-auth/AHAUCROTYJJY5ZSPIWVA2ODSMUJC3ANCNFSM4SMJY24Q .
Do you have to develop the transfer to Indian Accent?
I found I fell in love with Indian Accent somehow :)
Hi,
I have data on Indian english accent. Kindly suggest how i can use Kaldi using transfer learning.