Building the language model for fine tuning a model (transfer learning)


(Jahir) #1

I have trained a model on collected data. Now, I want to fine tune this model with my own voice data. My question is, should the language model for fine tuning contain only the transcripts of my own voice data. Or should it contain the transcripts from the previous model too?


(Murugan R) #2

@jahir sir i think you don’t need to create your own language model. deepspeech language model will adapt your data . (LM build by librispeech-lm-norm.txt.gz 5.5 GB txt file).

or if you build own lm then follow /data/lm/READme.txt.
it will helpful for you.