Romain Hennequin 231c164cd4 Merge pull request #375 from deezer/avoid_multiple_restore_librosa ...

Fix some issues with librosa stft backend: avoid multiple checkpoint restoration + one tf.graph per Separator. Version bumped to 1.5.2.

2020-05-15 16:11:45 +02:00

..

audio

Use tf.abs in compute_spectrogram_tf

2020-05-07 02:00:31 -07:00

commands

Removing duplicate logs

2020-03-20 15:39:11 +01:00

model

Remove temp model archive after extraction

2020-04-25 22:06:17 -07:00

resources

Added new config for processing until 16kHz

2019-12-27 14:34:27 +01:00

utils

Adding option to use librosa backend.

2020-02-26 16:59:21 +01:00

__init__.py

feat: add detailed ffprobe logs

2019-11-20 14:33:18 +01:00

__main__.py

feat: add detailed ffprobe logs

2019-11-20 14:33:18 +01:00

dataset.py

fix: remove unused dataset default path

2019-11-19 14:07:27 +01:00

separator.py

Avoid multiple checkpoint restauration and add a specific tf graph for each Separator which makes it possible to instanciate several ones

2020-05-12 21:48:41 +02:00