Delete spgm_train.yaml
Browse files- spgm_train.yaml +0 -176
spgm_train.yaml
DELETED
@@ -1,176 +0,0 @@
|
|
1 |
-
# ################################
|
2 |
-
# Model: SPGM for source separation
|
3 |
-
# https://arxiv.org/abs/2309.12608
|
4 |
-
# Dataset : WSJ0-2mix
|
5 |
-
# ################################
|
6 |
-
#
|
7 |
-
# Basic parameters
|
8 |
-
# Seed needs to be set at top of yaml, before objects with parameters are made
|
9 |
-
#
|
10 |
-
seed: 1234
|
11 |
-
__set_seed: !apply:torch.manual_seed [!ref <seed>]
|
12 |
-
|
13 |
-
# Data params
|
14 |
-
|
15 |
-
# e.g. '/yourpath/wsj0-mix/2speakers'
|
16 |
-
# end with 2speakers for wsj0-2mix or 3speakers for wsj0-3mix
|
17 |
-
data_folder:
|
18 |
-
|
19 |
-
# the path for wsj0/si_tr_s/ folder -- only needed if dynamic mixing is used
|
20 |
-
# e.g. /yourpath/wsj0-processed/si_tr_s/
|
21 |
-
# you need to convert the original wsj0 to 8k
|
22 |
-
# you can do this conversion with the script ../meta/preprocess_dynamic_mixing.py
|
23 |
-
base_folder_dm:
|
24 |
-
|
25 |
-
experiment_name: SPGM
|
26 |
-
output_folder: !ref results/<experiment_name>/<seed>
|
27 |
-
train_log: !ref <output_folder>/train_log.txt
|
28 |
-
save_folder: !ref <output_folder>/save
|
29 |
-
train_data: !ref <save_folder>/wsj_tr.csv
|
30 |
-
valid_data: !ref <save_folder>/wsj_cv.csv
|
31 |
-
test_data: !ref <save_folder>/wsj_tt.csv
|
32 |
-
skip_prep: False
|
33 |
-
|
34 |
-
ckpt_interval_minutes: 15
|
35 |
-
|
36 |
-
# Experiment params
|
37 |
-
auto_mix_prec: False # Set it to True for mixed precision
|
38 |
-
test_only: False
|
39 |
-
num_spks: 2 # set to 3 for wsj0-3mix
|
40 |
-
noprogressbar: False
|
41 |
-
save_audio: False # Save estimated sources on disk
|
42 |
-
sample_rate: 8000
|
43 |
-
|
44 |
-
# Training parameters
|
45 |
-
N_epochs: 200
|
46 |
-
batch_size: 1
|
47 |
-
lr: 0.00015
|
48 |
-
clip_grad_norm: 5
|
49 |
-
loss_upper_lim: 999999 # this is the upper limit for an acceptable loss
|
50 |
-
# if True, the training sequences are cut to a specified length
|
51 |
-
limit_training_signal_len: False
|
52 |
-
# this is the length of sequences if we choose to limit
|
53 |
-
# the signal length of training sequences
|
54 |
-
training_signal_len: 9999999999999999
|
55 |
-
|
56 |
-
# Set it to True to dynamically create mixtures at training time
|
57 |
-
dynamic_mixing: True
|
58 |
-
|
59 |
-
# Parameters for data augmentation
|
60 |
-
use_wavedrop: False
|
61 |
-
use_speedperturb: True
|
62 |
-
use_rand_shift: False
|
63 |
-
min_shift: -8000
|
64 |
-
max_shift: 8000
|
65 |
-
|
66 |
-
speedperturb: !new:speechbrain.lobes.augment.TimeDomainSpecAugment
|
67 |
-
perturb_prob: 1.0
|
68 |
-
drop_freq_prob: 0.0
|
69 |
-
drop_chunk_prob: 0.0
|
70 |
-
sample_rate: !ref <sample_rate>
|
71 |
-
speeds: [95, 100, 105]
|
72 |
-
|
73 |
-
wavedrop: !new:speechbrain.lobes.augment.TimeDomainSpecAugment
|
74 |
-
perturb_prob: 0.0
|
75 |
-
drop_freq_prob: 1.0
|
76 |
-
drop_chunk_prob: 1.0
|
77 |
-
sample_rate: !ref <sample_rate>
|
78 |
-
|
79 |
-
# loss thresholding -- this thresholds the training loss
|
80 |
-
threshold_byloss: True
|
81 |
-
threshold: -30
|
82 |
-
|
83 |
-
# Encoder parameters
|
84 |
-
N_encoder_out: 256
|
85 |
-
out_channels: 256
|
86 |
-
kernel_size: 16
|
87 |
-
kernel_stride: 8
|
88 |
-
|
89 |
-
# Dataloader options
|
90 |
-
# Set num_workers: 0 on MacOS due to behavior of the multiprocessing library
|
91 |
-
dataloader_opts:
|
92 |
-
batch_size: !ref <batch_size>
|
93 |
-
num_workers: 3
|
94 |
-
|
95 |
-
|
96 |
-
# Specifying the network
|
97 |
-
Encoder: !new:speechbrain.lobes.models.dual_path.Encoder
|
98 |
-
kernel_size: !ref <kernel_size>
|
99 |
-
out_channels: !ref <N_encoder_out>
|
100 |
-
|
101 |
-
|
102 |
-
SBtfintra: !new:speechbrain.lobes.models.dual_path.SBTransformerBlock
|
103 |
-
num_layers: 8
|
104 |
-
d_model: !ref <out_channels>
|
105 |
-
nhead: 8
|
106 |
-
d_ffn: 1024
|
107 |
-
dropout: 0
|
108 |
-
use_positional_encoding: True
|
109 |
-
norm_before: True
|
110 |
-
|
111 |
-
SBtfinter: !new:speechbrain.lobes.models.SPGM.SPGMBlock
|
112 |
-
n_embd: !ref <out_channels>
|
113 |
-
pool: 'att'
|
114 |
-
|
115 |
-
MaskNet: !new:speechbrain.lobes.models.dual_path.Dual_Path_Model
|
116 |
-
num_spks: !ref <num_spks>
|
117 |
-
in_channels: !ref <N_encoder_out>
|
118 |
-
out_channels: !ref <out_channels>
|
119 |
-
num_layers: 4
|
120 |
-
K: 250
|
121 |
-
intra_model: !ref <SBtfintra>
|
122 |
-
inter_model: !ref <SBtfinter>
|
123 |
-
norm: ln
|
124 |
-
linear_layer_after_inter_intra: False
|
125 |
-
skip_around_intra: True
|
126 |
-
|
127 |
-
Decoder: !new:speechbrain.lobes.models.dual_path.Decoder
|
128 |
-
in_channels: !ref <N_encoder_out>
|
129 |
-
out_channels: 1
|
130 |
-
kernel_size: !ref <kernel_size>
|
131 |
-
stride: !ref <kernel_stride>
|
132 |
-
bias: False
|
133 |
-
|
134 |
-
optimizer: !name:torch.optim.Adam
|
135 |
-
lr: !ref <lr>
|
136 |
-
weight_decay: 0
|
137 |
-
|
138 |
-
loss: !name:speechbrain.nnet.losses.get_si_snr_with_pitwrapper
|
139 |
-
|
140 |
-
lr_scheduler: !new:speechbrain.nnet.schedulers.ReduceLROnPlateau
|
141 |
-
factor: 0.5
|
142 |
-
patience: 2
|
143 |
-
dont_halve_until_epoch: 85
|
144 |
-
|
145 |
-
epoch_counter: !new:speechbrain.utils.epoch_loop.EpochCounter
|
146 |
-
limit: !ref <N_epochs>
|
147 |
-
|
148 |
-
modules:
|
149 |
-
encoder: !ref <Encoder>
|
150 |
-
decoder: !ref <Decoder>
|
151 |
-
masknet: !ref <MaskNet>
|
152 |
-
|
153 |
-
checkpointer: !new:speechbrain.utils.checkpoints.Checkpointer
|
154 |
-
checkpoints_dir: !ref <save_folder>
|
155 |
-
recoverables:
|
156 |
-
encoder: !ref <Encoder>
|
157 |
-
decoder: !ref <Decoder>
|
158 |
-
masknet: !ref <MaskNet>
|
159 |
-
counter: !ref <epoch_counter>
|
160 |
-
lr_scheduler: !ref <lr_scheduler>
|
161 |
-
|
162 |
-
train_logger: !new:speechbrain.utils.train_logger.FileTrainLogger
|
163 |
-
save_file: !ref <train_log>
|
164 |
-
|
165 |
-
|
166 |
-
# If you do not want to use the pretrained separator you can simply delete pretrained_separator field.
|
167 |
-
# pretrained_separator: !new:speechbrain.utils.parameter_transfer.Pretrainer
|
168 |
-
# collect_in: !ref <save_folder>
|
169 |
-
# loadables:
|
170 |
-
# encoder: !ref <Encoder>
|
171 |
-
# decoder: !ref <Decoder>
|
172 |
-
# masknet: !ref <MaskNet>
|
173 |
-
# paths:
|
174 |
-
# encoder: yipjiaqi/spgm/encoder.ckpt
|
175 |
-
# decoder: yipjiaqi/spgm/decoder.ckpt
|
176 |
-
# masknet: yipjiaqi/spgm/masknet.ckpt
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|