Is it possible to merge multiple acoustic models?

Archives DeepSpeech
1

Hi,
I had trained two sperate language models (lm1.binary and lm2.binary) from one alphabet source and two different vocabularies.
Separately they are working ok.
I’d like to merge them to avoid start the training over from the begining.
Is it possible?
Thanks

Edit: it’s about the acoustic model not the language model

2

If you produced them, then you can combine and produce a new one, I don’t understand where is the problem here.

3

Thanks for your reply in a short notice.
Training a model needs time, if we can merge two already trained models I will gain time.
Is it technically possible?

4

I’m not sure we are talking about the same thing, can you be more explicit ?

5

I have trained a model with a set of sentences/audiofiles, then I trained a second model with different set of sentences/audiofiles.
I was wondering if somehow I can merge the two models.

Am afraid I can’t be more explicit, else I can share training inputs?

6

Then you are not talking about language model but acoustic model, and no, we don’t have any solution to “merge” two models.

7

Sorry for not using the correct term that led to the confusion.

Do you know where can I find the description of the acoustic model?
I’d like to see if I can tweek the code to make it possible.

Any hint will be welcome.

8

The description ? In DeepSpeech.py ?