Anjok07 / ultimatevocalremovergui

GUI for a Vocal Remover that uses Deep Neural Networks.
MIT License
16.44k stars 1.24k forks source link

Fix "Audio buffer is not finite everywhere" by force initializing buffer for spec_s #1359

Open lspr98 opened 1 month ago

lspr98 commented 1 month ago

Initializing spec_s with np.ndarray can contain Nan / Inf according to the numpy docs, since it does not overwrite the underlying buffer memory. This causes the "Audio buffer is not finite everywhere" error in _cmb_spectrogram_towave, as already reported in multiple issues (#1021, #1045, #1147, #1153, #1262, to name a few). Initializing with np.zeros populates the buffer with zeros, avoiding the issue.