Upload melbandroformers2Fvoc_gabox.yaml

misc/melbandroformers2Fvoc_gabox.yaml

@@ -1,5 +1,5 @@
 audio:
-  chunk_size: 352800
+  chunk_size: 661500
   dim_f: 1024
   dim_t: 256
   hop_length: 441
@@ -22,7 +22,7 @@ model:
   ff_dropout: 0
   flash_attn: True
   dim_freqs_in: 1025
-  sample_rate: 44100  # needed for mel filter bank from librosa
+  sample_rate: 44100
   stft_n_fft: 2048
   stft_hop_length: 441
   stft_win_length: 2048
@@ -39,13 +39,16 @@ model:
   multi_stft_normalized: False
 
 training:
+  batch_size: 1
+  gradient_accumulation_steps: 1
+  grad_clip: 0
   instruments:
-  - Vocals
-  - Instrumental
-  target_instrument: Vocals
+  - vocals
+  - other
+  target_instrument: vocals
+  use_amp: true
 
 inference:
-  batch_size: 1
+  batch_size: 4
   dim_t: 1101
-  num_overlap: 1
-  chunk_size: 352800
+  num_overlap: 4