STT model deepspeech 0.7.1 results!

Dear Friends,

I am training a model with deepspeech 0.8.2 and afterwords, testing it. Maybe I am wrong, but I have feeling that even I replace the model and test with deepspeech alignmet 0.7.1 (current version), the result is same (old model = new more trained model) which was not likely for deepspeech 0.6.1. For this reason, I am confuse whether the loaded weights are getting saved at some place?
To be sure, can you please guide me, how to wash that memory and cross confirm that the replaced new model weights are loaded.
sorry for stupid question but this is not the case with alignment and transcription with ds 0.6.1 and if this is true, this can be leading to wrong outcome.

Please if you can guide me how to resolve this and check if the weights are not getting saved while testing any example.
Tools used by me are:
OS: CentOS
DS training: 0.8.2
DSAlign: 0.7.1
TF: 1.15.2
Training and align are tested on same machines, as well as, testing on different. Result is same.
Sorry if my finding is wrong. Need your opinion / guide.

@Tortoise I’m really sorry but I don’t understand the problem you are stating / facing.
There was no big change between 0.7 and 0.8, so if you get the same results, it’s more or less expected.

1 Like