I have a question sir. vocab.txt -> lm.arpa -> lm.binary & trie. this is the process for pretrained binaries. but i have my audio vocab.txt is there. but i can’t use or added into that frozen model retraining.
then how it will predict my audio technical terms or words.?
in my starting stage i run that model it is not predicting my technical term words. that the reason only i again retrain a frozen model.
if there is any possible ways is there to add my audio vocab.txt binaries , will it possible?
Well you should be able to augment the language model that way, yes. This is already extensively documented elsewhere on the forum. The only limitation right now is that I dont think https://github.com/mozilla/DeepSpeech/blob/master/data/lm/vocab.txt is actually the proper source file of https://github.com/mozilla/DeepSpeech/blob/master/data/lm/lm.binary
Maybe @kdavis might know if I’m mistaken?
@lissyx
Well you should be able to augment the language model that way
yes sir i already did this. i follow their instruction,
TUTORIAL : How I trained a specific french model to control my robot
i followed this sir instructions and then `
my own audio files vocab.txt.
alphabet.txt .
create a csv files like audio_file_dev.csv, audio_file_test.csv, audio_file_train.csv
i create my audio file vocab.txt --> lm.arpa --> lm.binary & trie
i run a DeepSpeech.py
it will generate a output_model.pb.
i test that output_model.pb, it not given a good results sir.
what i have to do ? if i will use my output_model.pb it not form speech sentence. it will display one or two technical term trained words only.
i was tried to implemented language model, but it is not working. because of amount of data is very less(50 audio files). so it is not gave a proper result.
so second i tried to implemented a continues fine tuning, it is only accept preddefined your deepspeech binary files(ex: lm.binary, trie). it is not predict my audio technical term words.
what i do sir? help me. thank you sir.
Ok, you get that when you built your own language model, and your own deepspeech model, right?
Can you document the source material for vocab.txt as well as your CSV files?
Was your deepspeech output_model.pb file produced from scratch or was it a fine-tuning one from our v0.1.1 model ?