I am trying to replicate this and it seems to be working on my end. Can you send the exact command-line you are using. Do the clips play in sfplay ? Can you replicate the problem using the datasets available in the Marsyas website ?
thanks,
George
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Yes, clips play in sfplay without problems, they also work on bextract for default sizes (512). Don't know if i can replicate the problem yet, need to finish the download of those packs first.
Anyway, here is the 24sec clip + mf i'm using (http://student.dei.uc.pt/~panda/marsyas/testcase1.zip).
The exact command i'm running is: "bextract.exe test.mf -fe -ws 10240 -hp 10240" and it crashes during feature extraction.
Sorry I should have caught this earlier. Currently the FFT calculation requires windows that are powers of two. There is functionality for zero-padding in Windowing but it is not connected in bextract so that any window size is supported. This should be clarified in the documentation and ideally fixed by appropriately zeropadding before the FFT calculation for the feature extraction
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Additional info: I'm using sound clips of 25 seconds in PCM WAV format (16bits, mono, 22050Hz).
I am trying to replicate this and it seems to be working on my end. Can you send the exact command-line you are using. Do the clips play in sfplay ? Can you replicate the problem using the datasets available in the Marsyas website ?
thanks,
George
Yes, clips play in sfplay without problems, they also work on bextract for default sizes (512). Don't know if i can replicate the problem yet, need to finish the download of those packs first.
Anyway, here is the 24sec clip + mf i'm using (http://student.dei.uc.pt/~panda/marsyas/testcase1.zip).
The exact command i'm running is: "bextract.exe test.mf -fe -ws 10240 -hp 10240" and it crashes during feature extraction.
It is indeed a strange bug because i can use a higher value without crashing it (20480) but as i described earlier in an email with this value it will crash if i call bextract_train_refactored(...) twice for the same collection. (I've described it in this message: https://sourceforge.net/mailarchive/message.php?msg_name=4C0D222B.2000501%40student.dei.uc.pt )
Hope it helps, could it be OS specific? (winXP here) Or a combination of settings/OS/file details?
Thanks a lot :)
Sorry I should have caught this earlier. Currently the FFT calculation requires windows that are powers of two. There is functionality for zero-padding in Windowing but it is not connected in bextract so that any window size is supported. This should be clarified in the documentation and ideally fixed by appropriately zeropadding before the FFT calculation for the feature extraction
You're right, tested with 8193 and 16384 and it works indeed. Strange it worked for some values like 20480 but i'm happy to finally know the cause :D