german-asr / german-asr-lm Goto Github PK

I am trying to make graph for german language which can be trained on Kaldi using any data like common voice or tuda or so.

I tried several ways which are still unsuccessful and everytime G.fst is empty.
Is it possible to use this to make model ? and train it _ any simple way of doing this ? because I found out your codes are way more easy to understand and user friendly.

if you can tell the way using own data to make LM for kaldi and then train with your script would be kind.

Earlier I learnt deepspeech from your codes . but for DS there is some limitation and it has signal overlap error of 10 ms.
Kaldi is much better.

Your helping reply is highly appreciated.

The release of pretrained models lists 6 different orders, could you add some advice on when to choose which? I am rather inexperienced with training models, so far only used a few.