Windows ASR toolkit based on Kaldi

[Daniel Povey]

This is an FYI: Zoltan Somongyi (cc'd) has created an ASR toolkit for Windows based on Kaldi, described here:

https://ai-toolkit.blogspot.be/p/voicebridge.html

Currently it only supports GMM-based ASR, and it has a different and simplified interface. It's possible that it may fill a useful niche.  If anyone wants to try it out and give comments to Zoltan (and myself as well, if you want), that would be great.  Particularly comments that would help improve VoiceBridge or Kaldi itself would be welcomed; and keep it positive, since he has done a lot of work on this.

Personally I don't have a Windows machine so I can't easily test it.

Dan

[Zoltán Somogyi]

Thank you for the kind introduction Dan! Everybody is welcome to test VoiceBridge. I will do my best to correct all problems you report as promptly as possible.

In order to extend what Dan said about the VoiceBridge models and for a bit more clarity the following models are currently available in VoiceBridge:

• Monophone,
  
• Delta + delta-delta triphone,
  
• LDA+MLLT,
  
• LDA+MLLT+SAT,
  
• DELTA+SAT (delta + delta-delta + SAT).
  

Currently the DELTA+SAT is the best performing model with the highest accuracy and speed (1/5th of the training time compared to LDA+MLLT+SAT). Due to the automatic tuning of some input (e.g. language model, pronunciation) VoiceBridge achieves the same accuracy in case of the clean LibriSpeech data as the DNN model in Kaldi! And this is exactly the aim of VoiceBridge!

Br,

Zoltan

[Xingyu Na]

Great work!

--
Go to http://kaldi-asr.org/forums.html find out how to join
---
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+unsubscribe@googlegroups.com.
To post to this group, send email to kaldi...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/kaldi-help/0b46217a-f0a5-483d-a588-5bb6bd4804a2%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--

Xingyu Na

[Zoltán Somogyi]

Thank you Xingyu Na!

I am looking forward for the input of the group for the further improvement of VoiceBridge! And of course all improvements will be applied to both Kaldi and VoiceBridge where appropriate! I share all info and knowledge with Dan.

On Monday, February 12, 2018 at 5:12:07 AM UTC+1, Xingyu Na wrote:

Great work!

2018-02-10 22:59 GMT+08:00 Zoltán Somogyi <zsomo...@gmail.com>:

Thank you for the kind introduction Dan! Everybody is welcome to test VoiceBridge. I will do my best to correct all problems you report as promptly as possible.

In order to extend what Dan said about the VoiceBridge models and for a bit more clarity the following models are currently available in VoiceBridge:

• Monophone,
  
• Delta + delta-delta triphone,
  
• LDA+MLLT,
  
• LDA+MLLT+SAT,
  
• DELTA+SAT (delta + delta-delta + SAT).
  

Currently the DELTA+SAT is the best performing model with the highest accuracy and speed (1/5th of the training time compared to LDA+MLLT+SAT). Due to the automatic tuning of some input (e.g. language model, pronunciation) VoiceBridge achieves the same accuracy in case of the clean LibriSpeech data as the DNN model in Kaldi! And this is exactly the aim of VoiceBridge!

Br,

Zoltan

On Friday, February 9, 2018 at 10:27:44 PM UTC+1, Dan Povey wrote:

This is an FYI: Zoltan Somongyi (cc'd) has created an ASR toolkit for Windows based on Kaldi, described here:

https://ai-toolkit.blogspot.be/p/voicebridge.html

Currently it only supports GMM-based ASR, and it has a different and simplified interface. It's possible that it may fill a useful niche.  If anyone wants to try it out and give comments to Zoltan (and myself as well, if you want), that would be great.  Particularly comments that would help improve VoiceBridge or Kaldi itself would be welcomed; and keep it positive, since he has done a lot of work on this.

Personally I don't have a Windows machine so I can't easily test it.

Dan

--
Go to http://kaldi-asr.org/forums.html find out how to join
---
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.

To post to this group, send email to kaldi...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/kaldi-help/0b46217a-f0a5-483d-a588-5bb6bd4804a2%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--

Xingyu Na

[Sarah Alyousefi]

Interesting,
Thank you for sharing