I am using Pocketsphinx to develop my ASR engine. I use the same wave files and sampling rate (16khz, mono) across different computers to run various experiments. However, I got the following in Phase 5:
"" Phase 5: Determine amount of training data, see if n_tied_states seems reasonable. Estimated Total Hours Training: 3.61094722222222 This is a small amount of data, no comment at this time""
However, using the same setup with the same wave files but on other computers, I got the following in Phase 5:
""Phase 5: Determine amount of training data, see if n_tied_states seems reasonable. Estimated Total Hours Training: 4.07851944444444 This is a small amount of data, no comment at this time""
I believe the correct estimated total hours training is 4.07851944444444 as stated in the second situation, because this number appeared in many other computers.
What could be the problem? same wave files, same parameters used for feature extraction, same sampling rate, ...etc.
Thank you
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I suspect you're using different versions of sphinxbase on different machines. up-to-date sphinxbase uses voice activity detector during feature extraction which filters silence hence there are less audio for training
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thank you very much for your reply. I actually used the same packages and tools (sphinxbase-0.8.tar; pocketsphinx-0.8.tar; sphinxtrain-1.0.8.tar) across all machines. Therefore, I don't think this is the reason
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hello,
I am using Pocketsphinx to develop my ASR engine. I use the same wave files and sampling rate (16khz, mono) across different computers to run various experiments. However, I got the following in Phase 5:
"" Phase 5: Determine amount of training data, see if n_tied_states seems reasonable. Estimated Total Hours Training: 3.61094722222222 This is a small amount of data, no comment at this time""
However, using the same setup with the same wave files but on other computers, I got the following in Phase 5:
""Phase 5: Determine amount of training data, see if n_tied_states seems reasonable. Estimated Total Hours Training: 4.07851944444444 This is a small amount of data, no comment at this time""
I believe the correct estimated total hours training is 4.07851944444444 as stated in the second situation, because this number appeared in many other computers.
What could be the problem? same wave files, same parameters used for feature extraction, same sampling rate, ...etc.
Thank you
I suspect you're using different versions of sphinxbase on different machines. up-to-date sphinxbase uses voice activity detector during feature extraction which filters silence hence there are less audio for training
Thank you very much for your reply. I actually used the same packages and tools (sphinxbase-0.8.tar; pocketsphinx-0.8.tar; sphinxtrain-1.0.8.tar) across all machines. Therefore, I don't think this is the reason