Gentle breaks sometimes

[smar10]

Hi, I'm trying to use Gentle , but sometimes it breaks without any error, just try to do the alignment "forever" without success. In some other audio files, works perfectly.

One example is this audio: motivation.wav.zip

With this words: "what keeps you motivated what keeps you motivated well i'm sure i'm here all my dreams and aspirations are here and the only thing in between these two is the bunch of work so tackling that work is a way of moving this needle"

[strob]

Does it with with the online demo? Are you using the latest release?

On Wed, Apr 19, 2017, 1:09 PM Sara Rijo notifications@github.com wrote:

Hi, I'm trying to use Gentle , but sometimes it breaks without any error, just try to do the alignment "forever" without success. In some other audio files, works perfectly.

One example is this audio: motivation.wav.zip https://github.com/lowerquality/gentle/files/936412/motivation.wav.zip

With this words: "what keeps you motivated what keeps you motivated well i'm sure i'm here all my dreams and aspirations are here and the only thing in between these two is the bunch of work so tackling that work is a way of moving this needle"

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/lowerquality/gentle/issues/134, or mute the thread https://github.com/notifications/unsubscribe-auth/AAMup5u0ThIj3EvLVJc6hW7I3Ur1MblGks5rxj-7gaJpZM4NB-ZT .

[smar10]

@strob I just run "docker run -P lowerquality/gentle" one week ago, i suppose it's the latest version. And yes, i'm using the graphical interface that docker container serves. The aligner works perfectly on some audio files/words , and fails sometimes in other. I can't discover any pattern (size of audio file, number of words) on failed alignments.

[strob]

It's probably a RAM issue. Try allocating more memory to the container. That said, the docker image isn't up to date, and should be updated.

On Wed, Apr 19, 2017, 1:38 PM Sara Rijo notifications@github.com wrote:

@strob https://github.com/strob I just run "docker run -P lowerquality/gentle" one week ago, i suppose it's the latest version. And yes, i'm using the graphical interface that docker container serves. The aligner works perfectly on some audio files/words , and fails sometimes in other. I can't discover any pattern (size of audio file, number of words) on failed alignments.

— You are receiving this because you were mentioned.

Reply to this email directly, view it on GitHub https://github.com/lowerquality/gentle/issues/134#issuecomment-295359835, or mute the thread https://github.com/notifications/unsubscribe-auth/AAMup7YLgTG885JQDKqA7aLvzM2xHVXzks5rxkZfgaJpZM4NB-ZT .

[smar10]

Thanks @strob , I will try !

[smar10]

@strob After updated the container and allocate more memory to the container, the error persist!

[piercefreeman]

+1 witnessing the same problem here with the command line utility (both on my own files and the provided motivation.wav.zip) - anyone have any ideas?

As a note, I'm running on a server with 200GB of RAM so unless Python/Kaldi have other caps that they impose internally, I don't think I'm hitting a limit here.

[strob]

Thanks for confirming this bug! I should hope Gentle doesn't eat through 200gb RAM quite yet... I've looked into it more closely, and pushed a temporary solution so that Gentle will no longer crash. Here is the transcription from motivation.wav.

There's a new problem in this example now, which is that the last word isn't being transcribed. A proper solution will involve figuring out why the nnet3 model fails on short chunks and/or making the segmentation algorithm more sophisticated. Reopening this issue until it is fully resolved.

Hope this helps for now!

[piercefreeman]

Hi @strob - Incredible, works like a charm now.

[smar10]

thanks @strob ! I will try that solution with other examples that failed in the past.