Resynthesis leads to corrupted output

[saikrishnarallabandi]

It seems to get corrupted on extraction and resynthesis
Music becomes noise

ADHBSU/aligned/wav/B01___01_Matthew_____ADHBSUN2DA_00005.wav

[saikrishnarallabandi]

@saikrishnarallabandi

[saikrishnarallabandi]

@nishantgurunath feel free to add more details

[saikrishnarallabandi]

@nishantgurunath cud confirm the issue.

Resynthesized file here:
http://tts.speech.cs.cmu.edu/rsk/misc_stuff/B01___01_Matthew_____ADHBSUN2DA_00005_reconstructed.wav

Looks like this is due to the spectral representation we are using. Let me look at other representations

[saikrishnarallabandi]

Extracting features at different frame shifts doesnt resolve the issue:

1 msec: http://tts.speech.cs.cmu.edu/rsk/misc_stuff/B01___01_Matthew_____ADHBSUN2DA_00005_reconstructed_1msec.wav

20 msec:
http://tts.speech.cs.cmu.edu/rsk/misc_stuff/B01___01_Matthew_____ADHBSUN2DA_00005_reconstructed_20msec.wav