Is latest models better than previous ones?

DeepSpeech
#1

I have a confusion about different versions of models available. I would like to explain my scenario first so that I can be more specific.

Initially, I took the pre-trained model with version 0.2.0a8 and trained with CVD(let’s call this model DS-2). Training took me 2 and half months. I have some more own data for further training which I have prepared. So I decided to train further as because till now I am not happy with the result I am getting.

here are my questions.

  1. I see many model versions for deepseech. Is the latest one better than the previous one with respect to performance? OR whats the main difference between difference models?

  2. For some reasons I decided to use 0.3.0 version now and train that model(although still not sure if I should do this). Is training DS-2 with additional data better or training 0.3.0 with CVD + my own data better?

  3. In case I decide to train 0.3.0, this time I want to use both the data(CVD + my own data). and I don’t want to wait long. So I am thinking of using google cloud ML for training the model. If any 1 has done this before, kindly give me some inputs.

Thanks :slight_smile:

#2

We try to stay at a similar level

Network changed, and I don’t think it’s compatible, so you cannot restart training, you need to re-do it from scratch.

Please refer to other threads on the forum about Google Cloud ML, we don’t use that.