how to build a gmm-hmm asr system?

[nerd cs]

I want to built a asr system based on gmm-hmm, but I don't know how to use gmm in kaldi,could anyone teach me something about this?Thanks very much!

[Danijel Korzinek]

Try opening some of the examples in the "egs" folder. You can start with Tedlium, Voxforge or Librispeech (if you have the resources) because they are all free. Most of the systems in the beginning (mono, tri1, tri2, etc) there are GMM based.

[nerd ha]

Thanks very much for your reply！

I know what you mean，some of the examples in the “egs”folder are used to build a dnn-hmm based system，and the mono，tri1 etc，are based on gmm，but what i meant was to build a gmm-hmm model，not dnn-hmm model

在 2016年11月29日星期二 UTC+8下午10:00:12，Danijel Korzinek写道：

[Daniel Povey]

The earlier stages, up to train_sat.sh and the associated decoding
stages, do that, i.e. it's based on gmm-hmm. Please do some
background reading on speech recognition before asking any more
questions, e.g. search for "gales young hmm" and read it.

> --
> You received this message because you are subscribed to the Google Groups
> "kaldi-help" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to kaldi-help+...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

[nerd ha]

Ok, thanks very much for your suggestion, I will read some relevant papers about these.

在 2016年11月30日星期三 UTC+8上午6:33:22，Dan Povey写道：