Open migueljette opened 7 years ago
Hi @strob any chance you can help me understand? From a Kaldi recipe perspective, which scripts should I run to get models to work within your framework? Thank you so much!!
or maybe @ronen has an idea. I just pulled your "dev" branch for gentle and it handles different "sample rate". So I imagine you figured out how to use different models with gentle. :) Thanks guys!
You need nnet2 online model, in librispeech recipe you need to reach this part
# # The following is the current online-nnet2 recipe, with "multi-splice".
# local/online/run_nnet2_ms.sh
then you simply replace the files in gentle models with the files of the same name from your model and you adjust the paths.
You will also have to update some sources, like this part must be changed:
config.silence_phones = "1:2:3:4:5:6:7:8:9:10:11:12:13:14:15:16:17:18:19:20";
@migueljette
or maybe @ronen has an idea. I just pulled your "dev" branch for gentle and it handles different "sample rate". So I imagine you figured out how to use different models with gentle. :)
Sorry I haven't... I made that change in anticipation of trying to figure it out, but haven't looked into it yet. In fact if you or @nshmyrev or @strob could point to some sort of instructions/help/tutorial for how to go about fiddling with or replacing the model that would be great. Thanks!
Hi, I'll try to figure it out now. I am running the run_nnet2_ms.sh recipe as I write this. Hopefully, it'll be done by next week and I can try this out with my own models. I'll write back if I make it work!
Hi there,
By the way, I was able to make this work. I had to update the folders "data" "PROTO_LANGDIR" with my own models (from run_nnet2_ms.sh)
I also had to update a few of the files in: "ext". Mostly "standard_kaldi.cc" where you have to give the name of a few of the files and also define which phonemes are silence phonemes. It would be nice to make this a bit more configurable in the future. If I have some time, I will make it so. For now, it works! This tool is awesome!
Hey, @migueljette -- this is thrilling!
I'd love for any tips you're willing to disclose. I would very much like to support a repository of different pre-compiled language models that people can use with Gentle.
I have a pull request to make it easier to use custom models here: https://github.com/lowerquality/gentle/pull/140
Hi, @migueljette. Could you show more details about the process of replacing custom models by steps?
Hi,
Do you have pointers on how to use Gentle with my own models? I have built Kaldi models with my own data (created my own recipe following Librispeech) and now I would like to test Gentle with my own model. Where do I get the files from PROTO_LANGDIR/ and data/?
Thanks for your help and direction!