You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I'm suspecting there's something weird going on with windowing or something, the output is always very heavily banded, like there's an amplitude modulation on the sound. Then again, it's maybe even frequency-dependent, so the frequency of the modulation seems to be slightly different at different frequencies.
I tried changing the FFT size and hop, but as you know, increasing it ouright resulted in an even more pronounced effect, and decreasing the hop proportionally gave a time compression like outcome. I now created a new model where I put a 0.2s sample overlap value I haven't tried it yet.
Otherwise the model seems very exciting, I made a 100k percussion model, and now it can reconstruct and vary things nicely except for that banding.
The text was updated successfully, but these errors were encountered:
I'm suspecting there's something weird going on with windowing or something, the output is always very heavily banded, like there's an amplitude modulation on the sound. Then again, it's maybe even frequency-dependent, so the frequency of the modulation seems to be slightly different at different frequencies.
Here's a few examples. The effect is very notable, but so fast that you'll need a short window size to see it:
https://ibb.co/XW8QW3c
https://ibb.co/T8MLxLz
I tried changing the FFT size and hop, but as you know, increasing it ouright resulted in an even more pronounced effect, and decreasing the hop proportionally gave a time compression like outcome. I now created a new model where I put a 0.2s sample overlap value I haven't tried it yet.
Otherwise the model seems very exciting, I made a 100k percussion model, and now it can reconstruct and vary things nicely except for that banding.
The text was updated successfully, but these errors were encountered: