-
Notifications
You must be signed in to change notification settings - Fork 35
Description
Have been trying many paramaters when generating songs.
even using the same parameters sometimes the outputs are very realistic,,,then the next generation the sound is totally silent.
Have no idea how to debug this. I will send as best i can the output of the generation:
when looking in the generated song in the player there appears to be 'sound bars' here and there but is totally silent.
`Inference started. Outputs will be saved in workspace/outputs...Starting stage 1...
Creating empty input seq
Section 1: output length 3194 exceeding context length 2199, now using the last 2199 tokens.
Section 2: output length 6054 exceeding context length 2199, now using the last 2199 tokens.
Starting stage 2...
x.max()=464, codebook_size=1024
x.max()=1013, codebook_size=1024
Starting postprocessing...
Processing workspace/outputs/stage2/vtrack.npy
Compressed shape: (8, 4200)
Decoded in 3.75s (22.37x RTF)
Saved: workspace/outputs/vocoder/stems/vtrack.mp3
Processing workspace/outputs/stage2/itrack.npy
Compressed shape: (8, 4200)
Decoded in 0.09s (916.41x RTF)
Saved: workspace/outputs/vocoder/stems/itrack.mp3
Created mix: workspace/outputs/vocoder/mix/mixed.mp3
Successfully created '1740346584_mixed.mp3' with matched low-frequency energy.
TIA
`