Spaces:
Runtime error
Runtime error
mattricesound
commited on
Commit
•
5945136
1
Parent(s):
106ab10
Tune effects, replace models with asteroid, remove silence
Browse files- cfg/effects/all.yaml +13 -7
- cfg/exp/dist.yaml +29 -0
- cfg/model/dcunet.yaml +3 -11
- cfg/model/dptnet.yaml +10 -6
- remfx/models.py +5 -4
- setup.py +1 -0
cfg/effects/all.yaml
CHANGED
@@ -4,13 +4,19 @@ effects:
|
|
4 |
chorus:
|
5 |
_target_: remfx.effects.RandomPedalboardChorus
|
6 |
sample_rate: ${sample_rate}
|
|
|
|
|
|
|
|
|
7 |
min_depth: 0.2
|
8 |
-
|
|
|
|
|
9 |
distortion:
|
10 |
_target_: remfx.effects.RandomPedalboardDistortion
|
11 |
sample_rate: ${sample_rate}
|
12 |
-
min_drive_db:
|
13 |
-
max_drive_db:
|
14 |
compressor:
|
15 |
_target_: remfx.effects.RandomPedalboardCompressor
|
16 |
sample_rate: ${sample_rate}
|
@@ -26,7 +32,7 @@ effects:
|
|
26 |
min_damping: 0.2
|
27 |
max_damping: 1.0
|
28 |
min_wet_dry: 0.2
|
29 |
-
max_wet_dry: 0.
|
30 |
min_width: 0.2
|
31 |
max_width: 1.0
|
32 |
delay:
|
@@ -35,6 +41,6 @@ effects:
|
|
35 |
min_delay_seconds: 0.1
|
36 |
max_delay_sconds: 1.0
|
37 |
min_feedback: 0.05
|
38 |
-
max_feedback: 0.
|
39 |
-
min_mix: 0.
|
40 |
-
max_mix: 0.
|
|
|
4 |
chorus:
|
5 |
_target_: remfx.effects.RandomPedalboardChorus
|
6 |
sample_rate: ${sample_rate}
|
7 |
+
min_rate_hz: 0.25
|
8 |
+
max_rate_hz: 1.5
|
9 |
+
min_feedback: 0.1
|
10 |
+
max_feedback: 0.4
|
11 |
min_depth: 0.2
|
12 |
+
max_depth: 0.6
|
13 |
+
min_mix: 0.15
|
14 |
+
max_mix: 0.4
|
15 |
distortion:
|
16 |
_target_: remfx.effects.RandomPedalboardDistortion
|
17 |
sample_rate: ${sample_rate}
|
18 |
+
min_drive_db: 7
|
19 |
+
max_drive_db: 25
|
20 |
compressor:
|
21 |
_target_: remfx.effects.RandomPedalboardCompressor
|
22 |
sample_rate: ${sample_rate}
|
|
|
32 |
min_damping: 0.2
|
33 |
max_damping: 1.0
|
34 |
min_wet_dry: 0.2
|
35 |
+
max_wet_dry: 0.6
|
36 |
min_width: 0.2
|
37 |
max_width: 1.0
|
38 |
delay:
|
|
|
41 |
min_delay_seconds: 0.1
|
42 |
max_delay_sconds: 1.0
|
43 |
min_feedback: 0.05
|
44 |
+
max_feedback: 0.3
|
45 |
+
min_mix: 0.1
|
46 |
+
max_mix: 0.35
|
cfg/exp/dist.yaml
ADDED
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# @package _global_
|
2 |
+
defaults:
|
3 |
+
- override /model: umx
|
4 |
+
- override /effects: all
|
5 |
+
seed: 12345
|
6 |
+
sample_rate: 48000
|
7 |
+
chunk_size: 262144 # 5.5s
|
8 |
+
logs_dir: "./logs"
|
9 |
+
render_files: True
|
10 |
+
render_root: "./data"
|
11 |
+
accelerator: "gpu"
|
12 |
+
log_audio: True
|
13 |
+
# Effects
|
14 |
+
max_kept_effects: 5
|
15 |
+
max_removed_effects: -1
|
16 |
+
shuffle_kept_effects: True
|
17 |
+
shuffle_removed_effects: False
|
18 |
+
num_classes: 5
|
19 |
+
effects_to_use:
|
20 |
+
- compressor
|
21 |
+
- distortion
|
22 |
+
- reverb
|
23 |
+
- chorus
|
24 |
+
- delay
|
25 |
+
effects_to_remove:
|
26 |
+
- distortion
|
27 |
+
datamodule:
|
28 |
+
batch_size: 16
|
29 |
+
num_workers: 8
|
cfg/model/dcunet.yaml
CHANGED
@@ -9,16 +9,8 @@ model:
|
|
9 |
sample_rate: ${sample_rate}
|
10 |
network:
|
11 |
_target_: remfx.models.DCUNetModel
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
hop_len: 64
|
16 |
-
block_layers: 4
|
17 |
-
layers: 4
|
18 |
-
kernel_size: 3
|
19 |
-
refine_layers: 1
|
20 |
-
is_mask: True
|
21 |
-
norm: 'ins'
|
22 |
-
act: 'comp'
|
23 |
sample_rate: ${sample_rate}
|
24 |
num_bins: 1025
|
|
|
9 |
sample_rate: ${sample_rate}
|
10 |
network:
|
11 |
_target_: remfx.models.DCUNetModel
|
12 |
+
architecture: "DCUNet-10"
|
13 |
+
stft_kernel_size: 512
|
14 |
+
fix_length_mode: "pad"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
sample_rate: ${sample_rate}
|
16 |
num_bins: 1025
|
cfg/model/dptnet.yaml
CHANGED
@@ -9,12 +9,16 @@ model:
|
|
9 |
sample_rate: ${sample_rate}
|
10 |
network:
|
11 |
_target_: remfx.models.DPTNetModel
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
|
|
18 |
win_len: 2
|
|
|
|
|
|
|
19 |
sample_rate: ${sample_rate}
|
20 |
num_bins: 1025
|
|
|
9 |
sample_rate: ${sample_rate}
|
10 |
network:
|
11 |
_target_: remfx.models.DPTNetModel
|
12 |
+
n_src: 1
|
13 |
+
in_chan: 1
|
14 |
+
out_chan: 1
|
15 |
+
bn_chan: 128
|
16 |
+
hid_size: 128
|
17 |
+
chunk_size: 100
|
18 |
+
n_repeats: 6
|
19 |
win_len: 2
|
20 |
+
fb_name: "free"
|
21 |
+
kernel_size: 16
|
22 |
+
n_filters: 1
|
23 |
sample_rate: ${sample_rate}
|
24 |
num_bins: 1025
|
remfx/models.py
CHANGED
@@ -13,6 +13,7 @@ from remfx.dptnet import DPTNet_base
|
|
13 |
from remfx.dcunet import RefineSpectrogramUnet
|
14 |
from remfx.tcn import TCN
|
15 |
from remfx.utils import causal_crop
|
|
|
16 |
|
17 |
|
18 |
class RemFX(pl.LightningModule):
|
@@ -197,7 +198,7 @@ class DiffusionGenerationModel(nn.Module):
|
|
197 |
class DPTNetModel(nn.Module):
|
198 |
def __init__(self, sample_rate, num_bins, **kwargs):
|
199 |
super().__init__()
|
200 |
-
self.model =
|
201 |
self.num_bins = num_bins
|
202 |
self.mrstftloss = MultiResolutionSTFTLoss(
|
203 |
n_bins=self.num_bins, sample_rate=sample_rate
|
@@ -217,7 +218,7 @@ class DPTNetModel(nn.Module):
|
|
217 |
class DCUNetModel(nn.Module):
|
218 |
def __init__(self, sample_rate, num_bins, **kwargs):
|
219 |
super().__init__()
|
220 |
-
self.model =
|
221 |
self.mrstftloss = MultiResolutionSTFTLoss(
|
222 |
n_bins=num_bins, sample_rate=sample_rate
|
223 |
)
|
@@ -225,7 +226,7 @@ class DCUNetModel(nn.Module):
|
|
225 |
|
226 |
def forward(self, batch):
|
227 |
x, target = batch
|
228 |
-
output = self.model(x.squeeze(1))
|
229 |
# Crop target to match output
|
230 |
if output.shape[-1] < target.shape[-1]:
|
231 |
target = causal_crop(target, output.shape[-1])
|
@@ -233,7 +234,7 @@ class DCUNetModel(nn.Module):
|
|
233 |
return loss, output
|
234 |
|
235 |
def sample(self, x: Tensor) -> Tensor:
|
236 |
-
output = self.model(x.squeeze(1))
|
237 |
return output
|
238 |
|
239 |
|
|
|
13 |
from remfx.dcunet import RefineSpectrogramUnet
|
14 |
from remfx.tcn import TCN
|
15 |
from remfx.utils import causal_crop
|
16 |
+
import asteroid
|
17 |
|
18 |
|
19 |
class RemFX(pl.LightningModule):
|
|
|
198 |
class DPTNetModel(nn.Module):
|
199 |
def __init__(self, sample_rate, num_bins, **kwargs):
|
200 |
super().__init__()
|
201 |
+
self.model = asteroid.models.dptnet.DPTNet(**kwargs)
|
202 |
self.num_bins = num_bins
|
203 |
self.mrstftloss = MultiResolutionSTFTLoss(
|
204 |
n_bins=self.num_bins, sample_rate=sample_rate
|
|
|
218 |
class DCUNetModel(nn.Module):
|
219 |
def __init__(self, sample_rate, num_bins, **kwargs):
|
220 |
super().__init__()
|
221 |
+
self.model = asteroid.models.DCUNet(**kwargs)
|
222 |
self.mrstftloss = MultiResolutionSTFTLoss(
|
223 |
n_bins=num_bins, sample_rate=sample_rate
|
224 |
)
|
|
|
226 |
|
227 |
def forward(self, batch):
|
228 |
x, target = batch
|
229 |
+
output = self.model(x.squeeze(1)) # B x 1 x T
|
230 |
# Crop target to match output
|
231 |
if output.shape[-1] < target.shape[-1]:
|
232 |
target = causal_crop(target, output.shape[-1])
|
|
|
234 |
return loss, output
|
235 |
|
236 |
def sample(self, x: Tensor) -> Tensor:
|
237 |
+
output = self.model(x.squeeze(1)) # B x 1 x T
|
238 |
return output
|
239 |
|
240 |
|
setup.py
CHANGED
@@ -48,6 +48,7 @@ setup(
|
|
48 |
"pedalboard",
|
49 |
"frechet_audio_distance",
|
50 |
"ordered-set",
|
|
|
51 |
],
|
52 |
include_package_data=True,
|
53 |
license="Apache License 2.0",
|
|
|
48 |
"pedalboard",
|
49 |
"frechet_audio_distance",
|
50 |
"ordered-set",
|
51 |
+
"asteroid",
|
52 |
],
|
53 |
include_package_data=True,
|
54 |
license="Apache License 2.0",
|