cemsubakan
commited on
Commit
•
ffdc70a
1
Parent(s):
f2a1ac3
Update hyperparams.yaml
Browse files- hyperparams.yaml +12 -30
hyperparams.yaml
CHANGED
@@ -1,15 +1,8 @@
|
|
1 |
-
|
2 |
-
# /home/mila/s/subakany/speechbrain-soundskrit/recipes/WHAMandWHAMR/enhancement/hparams/sepformer-whamr-DM.yaml
|
3 |
-
# yamllint disable
|
4 |
# ################################
|
5 |
# Model: SepFormer for source separation
|
6 |
-
# https://arxiv.org/abs/2010.13154
|
7 |
-
#
|
8 |
# Dataset : WHAMR!
|
9 |
# ################################
|
10 |
-
# Basic parameters
|
11 |
-
# Seed needs to be set at top of yaml, before objects with parameters are made
|
12 |
-
#
|
13 |
|
14 |
sample_rate: 8000
|
15 |
num_spks: 1
|
@@ -21,12 +14,11 @@ kernel_size: 16
|
|
21 |
kernel_stride: 8
|
22 |
|
23 |
# Specifying the network
|
24 |
-
Encoder:
|
25 |
kernel_size: 16
|
26 |
out_channels: 256
|
27 |
|
28 |
-
|
29 |
-
SBtfintra: &id001 !new:speechbrain.lobes.models.dual_path.SBTransformerBlock
|
30 |
num_layers: 8
|
31 |
d_model: 256
|
32 |
nhead: 8
|
@@ -35,7 +27,7 @@ SBtfintra: &id001 !new:speechbrain.lobes.models.dual_path.SBTransformerBlock
|
|
35 |
use_positional_encoding: true
|
36 |
norm_before: true
|
37 |
|
38 |
-
SBtfinter:
|
39 |
num_layers: 8
|
40 |
d_model: 256
|
41 |
nhead: 8
|
@@ -44,42 +36,32 @@ SBtfinter: &id002 !new:speechbrain.lobes.models.dual_path.SBTransformerBlock
|
|
44 |
use_positional_encoding: true
|
45 |
norm_before: true
|
46 |
|
47 |
-
MaskNet:
|
48 |
-
|
49 |
num_spks: 1
|
50 |
in_channels: 256
|
51 |
out_channels: 256
|
52 |
num_layers: 2
|
53 |
K: 250
|
54 |
-
intra_model:
|
55 |
-
inter_model:
|
56 |
norm: ln
|
57 |
linear_layer_after_inter_intra: false
|
58 |
skip_around_intra: true
|
59 |
|
60 |
-
Decoder:
|
61 |
in_channels: 256
|
62 |
out_channels: 1
|
63 |
kernel_size: 16
|
64 |
stride: 8
|
65 |
bias: false
|
66 |
|
67 |
-
|
68 |
modules:
|
69 |
-
encoder:
|
70 |
-
decoder:
|
71 |
-
masknet:
|
72 |
-
save_all_checkpoints: false
|
73 |
-
checkpointer: !new:speechbrain.utils.checkpoints.Checkpointer
|
74 |
-
checkpoints_dir: results/sepformer-whamr-enhancement-DM/1234/save
|
75 |
-
recoverables:
|
76 |
-
encoder: *id003
|
77 |
-
decoder: *id004
|
78 |
-
masknet: *id005
|
79 |
|
80 |
pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer
|
81 |
loadables:
|
82 |
encoder: !ref <Encoder>
|
83 |
masknet: !ref <MaskNet>
|
84 |
-
decoder: !ref <Decoder>
|
85 |
-
|
|
|
1 |
+
|
|
|
|
|
2 |
# ################################
|
3 |
# Model: SepFormer for source separation
|
|
|
|
|
4 |
# Dataset : WHAMR!
|
5 |
# ################################
|
|
|
|
|
|
|
6 |
|
7 |
sample_rate: 8000
|
8 |
num_spks: 1
|
|
|
14 |
kernel_stride: 8
|
15 |
|
16 |
# Specifying the network
|
17 |
+
Encoder: !new:speechbrain.lobes.models.dual_path.Encoder
|
18 |
kernel_size: 16
|
19 |
out_channels: 256
|
20 |
|
21 |
+
SBtfintra: !new:speechbrain.lobes.models.dual_path.SBTransformerBlock
|
|
|
22 |
num_layers: 8
|
23 |
d_model: 256
|
24 |
nhead: 8
|
|
|
27 |
use_positional_encoding: true
|
28 |
norm_before: true
|
29 |
|
30 |
+
SBtfinter: !new:speechbrain.lobes.models.dual_path.SBTransformerBlock
|
31 |
num_layers: 8
|
32 |
d_model: 256
|
33 |
nhead: 8
|
|
|
36 |
use_positional_encoding: true
|
37 |
norm_before: true
|
38 |
|
39 |
+
MaskNet: !new:speechbrain.lobes.models.dual_path.Dual_Path_Model
|
|
|
40 |
num_spks: 1
|
41 |
in_channels: 256
|
42 |
out_channels: 256
|
43 |
num_layers: 2
|
44 |
K: 250
|
45 |
+
intra_model: !ref <SBtfintra>
|
46 |
+
inter_model: !ref <SBtfinter>
|
47 |
norm: ln
|
48 |
linear_layer_after_inter_intra: false
|
49 |
skip_around_intra: true
|
50 |
|
51 |
+
Decoder: !new:speechbrain.lobes.models.dual_path.Decoder
|
52 |
in_channels: 256
|
53 |
out_channels: 1
|
54 |
kernel_size: 16
|
55 |
stride: 8
|
56 |
bias: false
|
57 |
|
|
|
58 |
modules:
|
59 |
+
encoder: !ref <Encoder>
|
60 |
+
decoder: !ref <Decoder>
|
61 |
+
masknet: !ref <MaskNet>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
62 |
|
63 |
pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer
|
64 |
loadables:
|
65 |
encoder: !ref <Encoder>
|
66 |
masknet: !ref <MaskNet>
|
67 |
+
decoder: !ref <Decoder>
|
|