Hi Nickolay sir, I was training the model earlier on windows but as you told some days before to try it on linux. Now when I am using all the files created on windows on the Ubuntu then it is not creating .mfc files i.e. it creates empty feat folder due to which training cannot proceed further. Can you please tell me some solution to this. Any reason for this. I can share all the files created if you will say. But please help me.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hello Sir, a large part of my training is complete, some modules completed fine but it is giving some errors in some modules in Training context dependent models phase. When I check the log files even then it seems fine as all the results have been written properly to the output files. So can you please help me in finding out the cause of errors. I shall be really very thankful to you for this. Here is the link of the log files and other related folders. https://drive.google.com/file/d/0Bw_kmpUVdN3ValdRVGdCam50TGc/view?usp=sharing
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Log says you're training on windows. Why you mentioned ubuntu? Your training is finished. Seems some audio files do not match tutorial suggestions: from 5 to 30 secs duration, 0.5 sec initial silence. Those files are ignored during training because of that ("Failed to align" errors). That's actually not a big deal, but you might want to fix it checking what's wrong with those files. During decoding stage, some mfcc occurred to be empty. That can be caused by VAD that filtered all the audio from concrete files as silence. Run decoding with "-remove_silence no" or fix files to match mentioned suggestions. Overall amount of data you're using too small, any usable model isn't expected
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Than you sir for this help. And the feat problem(mentioned by me earlier) continued on ubuntu, so I again tried it on windows which worked now. And the amount of data I am using is 1 hour recording. So according to you what could be the minimum size to get a good model?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi Nickolay sir, I was training the model earlier on windows but as you told some days before to try it on linux. Now when I am using all the files created on windows on the Ubuntu then it is not creating .mfc files i.e. it creates empty feat folder due to which training cannot proceed further. Can you please tell me some solution to this. Any reason for this. I can share all the files created if you will say. But please help me.
Logs should contain info on that
Hello Sir, a large part of my training is complete, some modules completed fine but it is giving some errors in some modules in Training context dependent models phase. When I check the log files even then it seems fine as all the results have been written properly to the output files. So can you please help me in finding out the cause of errors. I shall be really very thankful to you for this. Here is the link of the log files and other related folders.
https://drive.google.com/file/d/0Bw_kmpUVdN3ValdRVGdCam50TGc/view?usp=sharing
Log says you're training on windows. Why you mentioned ubuntu? Your training is finished. Seems some audio files do not match tutorial suggestions: from 5 to 30 secs duration, 0.5 sec initial silence. Those files are ignored during training because of that ("Failed to align" errors). That's actually not a big deal, but you might want to fix it checking what's wrong with those files. During decoding stage, some mfcc occurred to be empty. That can be caused by VAD that filtered all the audio from concrete files as silence. Run decoding with "-remove_silence no" or fix files to match mentioned suggestions. Overall amount of data you're using too small, any usable model isn't expected
Than you sir for this help. And the feat problem(mentioned by me earlier) continued on ubuntu, so I again tried it on windows which worked now. And the amount of data I am using is 1 hour recording. So according to you what could be the minimum size to get a good model?
This is covered in tutorial