I had initially ran training using cont and got 78% accuracy and as such decided to train in PTM to get better accuracy as suggested in the training guide moresothat I want to use my acoustic model in a mobile phone application (Android). However, upon training in PTM i get these errors which never reflected in cont as shown in the attached logdir:
1.Must be at least one line in the control file
2.Feature should be either .semi. or .cont., is .ptm.
3.The following senones never occur in the input data
Which results in "Training failed at iteration 1". Lastly, I am using 4700 as the test set of 46589 training set (1/10) as suggested from the training guide, however, in the training part some transcriptions are repeating. Is this a good number for the test set cause I have realized it plays a big role in the resulting accuracy? Please advice!
Hi,
I had initially ran training using cont and got 78% accuracy and as such decided to train in PTM to get better accuracy as suggested in the training guide moresothat I want to use my acoustic model in a mobile phone application (Android). However, upon training in PTM i get these errors which never reflected in cont as shown in the attached logdir:
1.Must be at least one line in the control file
2.Feature should be either .semi. or .cont., is .ptm.
3.The following senones never occur in the input data
Which results in "Training failed at iteration 1". Lastly, I am using 4700 as the test set of 46589 training set (1/10) as suggested from the training guide, however, in the training part some transcriptions are repeating. Is this a good number for the test set cause I have realized it plays a big role in the resulting accuracy? Please advice!
https://www.dropbox.com/s/ujt2fsafunk8wns/logdir.zip?dl=0