Upload 2 files
Browse files- apollo_model.ckpt +3 -0
- config_apollo_vocal.yaml +41 -0
apollo_model.ckpt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:61b93332da9cde13925b2726b0b31167b05bc20030271b19a583ccf76d69354b
|
| 3 |
+
size 66532080
|
config_apollo_vocal.yaml
ADDED
|
@@ -0,0 +1,41 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
audio:
|
| 2 |
+
chunk_size: 132300 #132300
|
| 3 |
+
num_channels: 2
|
| 4 |
+
sample_rate: 44100
|
| 5 |
+
min_mean_abs: 0.0
|
| 6 |
+
|
| 7 |
+
model:
|
| 8 |
+
sr: 44100
|
| 9 |
+
win: 20
|
| 10 |
+
feature_dim: 256
|
| 11 |
+
layer: 6
|
| 12 |
+
|
| 13 |
+
training:
|
| 14 |
+
instruments: ['restored', 'addition']
|
| 15 |
+
target_instrument: 'restored'
|
| 16 |
+
batch_size: 1
|
| 17 |
+
gradient_accumulation_steps: 2
|
| 18 |
+
num_steps: 1000
|
| 19 |
+
num_epochs: 1000
|
| 20 |
+
optimizer: 'prodigy'
|
| 21 |
+
lr: 1.0
|
| 22 |
+
patience: 2
|
| 23 |
+
reduce_factor: 0.95
|
| 24 |
+
coarse_loss_clip: true
|
| 25 |
+
grad_clip: 0
|
| 26 |
+
q: 0.95
|
| 27 |
+
use_amp: false
|
| 28 |
+
|
| 29 |
+
optimizer:
|
| 30 |
+
weight_decay: 0.01
|
| 31 |
+
betas: [0.5, 0.99]
|
| 32 |
+
|
| 33 |
+
augmentations:
|
| 34 |
+
enable: true # enable or disable all augmentations (to fast disable if needed)
|
| 35 |
+
loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max)
|
| 36 |
+
loudness_min: 0.5
|
| 37 |
+
loudness_max: 1.5
|
| 38 |
+
|
| 39 |
+
inference:
|
| 40 |
+
batch_size: 4
|
| 41 |
+
num_overlap: 4
|