Closed turian closed 3 years ago
I am invoking openl3 on one second audio. Centering is enabled:
openl3 1secondsounds/ --content-type music --input-repr mel256 --audio-embedding-size 512
This should pad with 0.5 seconds of silence at the beginning, which means I should have timestamps at 0.0, 0.1, ... 0.5, correct?
Instead I only have (1, 512) embedding and timestamp only 0.0.
Why?
Retracted because of user error. The corpus README said 1.0 seconds, but many files were shorter. With padding to 1.0 second, I get the expected result.
I am invoking openl3 on one second audio. Centering is enabled:
This should pad with 0.5 seconds of silence at the beginning, which means I should have timestamps at 0.0, 0.1, ... 0.5, correct?
Instead I only have (1, 512) embedding and timestamp only 0.0.
Why?