Now I developing Urudu SST(speech to text) project and I am using Kaldi ASPIRE recipe.
And I have 250hours corpus.
To making a model, I followed Kaldi manual steps but TDNN training was not converged.
I can't find the reason.
I think it can resolve by only a real-expert with Kaldi and speech-recognition skills.
This task is very urgent one for me.
If you are a real-expert and can help me in time, I will pay fully.
Please don't hesitate to bid if you are a good kaldi and ML expert.
Thanks.