Releases: bbc/audio-dafx2019-automatic
Releases · bbc/audio-dafx2019-automatic
Music/Speech/SFx discriminator model trained on GTZAN and BBC SFx
Fixed some bugs on the training code and trained on the CPU using a number of 4 epochs. On the same dataset it gives much higher f1 scores than before:
Type | precision | recall | f1-score | support |
---|---|---|---|---|
Music | 1.00 | 0.97 | 0.99 | 713 |
Speech | 0.98 | 0.98 | 0.98 | 527 |
SFx | 0.96 | 0.99 | 0.98 | 494 |
Music/Speech/SFx discriminator model trained on GTZAN and BBC SFx
.h5
and .json
files.
Music/Speech/SFx discriminator model trained on GTZAN
Adapted VGGish trained on GTZAN music/speech discriminator dataset and augmented by samples from the BBC SFx library