-
Notifications
You must be signed in to change notification settings - Fork 53
/
transformer_cpc1.yaml
268 lines (230 loc) · 8.18 KB
/
transformer_cpc1.yaml
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
# ############################################################################
# Model: E2E ASR with Transformer
# Encoder: Transformer Encoder
# Decoder: Transformer Decoder + (CTC/ATT joint) beamsearch + TransformerLM
# Tokens: unigram
# losses: CTC + KLdiv (Label Smoothing loss)
# Training: Librispeech 960h
# Authors: Jianyuan Zhong, Titouan Parcollet
# ############################################################################
# Seed needs to be set at top of yaml, before objects with parameters are made
seed: 8886
__set_seed: !apply:torch.manual_seed [!ref <seed>]
output_folder: !ref ???
wer_file: !ref <output_folder>/wer.txt
save_folder: !ref <output_folder>/save
train_log: !ref <output_folder>/train_log.txt
# Language model (LM) pretraining
# NB: To avoid mismatch, the speech recognizer must be trained with the same
# tokenizer used for LM training. Here, we download everything from the
# speechbrain HuggingFace repository. However, a local path pointing to a
# directory containing the lm.ckpt and tokenizer.ckpt may also be specified
# instead. E.g if you want to use your own LM / tokenizer.
pretrained_lm_tokenizer_path: speechbrain/asr-transformer-transformerlm-librispeech
# Data files
data_folder: !ref ???
train_splits: ["cpc1_train_binaural"]
dev_splits: ["cpc1_dev_binaural"]
test_splits: ["cpc1_train_left_dev", "cpc1_train_right_dev"]
skip_prep: True
train_csv: !ref <data_folder>/binaural_train_msbg.csv
valid_csv: !ref <data_folder>/binaural_dev_msbg.csv
test_csv:
- !ref <data_folder>/left_dev_msbg.csv
- !ref <data_folder>/right_dev_msbg.csv
ckpt_interval_minutes: 30 # save checkpoint every N min
# Training parameters
# To make Transformers converge, the global bath size should be large enough.
# The global batch size is computed as batch_size * n_gpus * gradient_accumulation.
# Empirically, we found that this value should be >= 128.
# Please, set your parameters accordingly.
number_of_epochs: 120
batch_size: 16
ctc_weight: 0.3
gradient_accumulation: 4
gradient_clipping: 5.0
loss_reduction: 'batchmean'
sorting: random
# stages related parameters
stage_one_epochs: 90
lr_adam: 1.0
lr_sgd: 0.000025
# Feature parameters
sample_rate: 16000
n_fft: 400
n_mels: 80
# Dataloader options
train_dataloader_opts:
batch_size: !ref <batch_size>
shuffle: True
valid_dataloader_opts:
batch_size: 1
test_dataloader_opts:
batch_size: 1
####################### Model parameters ###########################
# Transformer
d_model: 768
nhead: 8
num_encoder_layers: 12
num_decoder_layers: 6
d_ffn: 3072
transformer_dropout: 0.0
activation: !name:torch.nn.GELU
output_neurons: 5000
vocab_size: 5000
# Outputs
blank_index: 0
label_smoothing: 0.1
pad_index: 0
bos_index: 1
eos_index: 2
unk_index: 0
# Decoding parameters
min_decode_ratio: 0.0
max_decode_ratio: 1.0
valid_search_interval: 10
valid_beam_size: 10
test_beam_size: 66
lm_weight: 0.00
ctc_weight_decode: 0.40
############################## models ################################
CNN: !new:speechbrain.lobes.models.convolution.ConvolutionFrontEnd
input_shape: (8, 10, 80)
num_blocks: 3
num_layers_per_block: 1
out_channels: (128, 256, 512)
kernel_sizes: (3, 3, 1)
strides: (2, 2, 1)
residuals: (False, False, False)
Transformer: !new:speechbrain.lobes.models.transformer.TransformerASR.TransformerASR # yamllint disable-line rule:line-length
input_size: 10240
tgt_vocab: !ref <output_neurons>
d_model: !ref <d_model>
nhead: !ref <nhead>
num_encoder_layers: !ref <num_encoder_layers>
num_decoder_layers: !ref <num_decoder_layers>
d_ffn: !ref <d_ffn>
dropout: !ref <transformer_dropout>
activation: !ref <activation>
normalize_before: False
# This is the TransformerLM that is used according to the Huggingface repository
# Visit the HuggingFace model corresponding to the pretrained_lm_tokenizer_path
# For more details about the model!
# NB: It has to match the pre-trained TransformerLM!!
lm_model: !new:speechbrain.lobes.models.transformer.TransformerLM.TransformerLM # yamllint disable-line rule:line-length
vocab: !ref <output_neurons>
d_model: 768
nhead: 12
num_encoder_layers: 12
num_decoder_layers: 0
d_ffn: 3072
dropout: 0.0
activation: !name:torch.nn.GELU
normalize_before: False
tokenizer: !new:sentencepiece.SentencePieceProcessor
ctc_lin: !new:speechbrain.nnet.linear.Linear
input_size: !ref <d_model>
n_neurons: !ref <output_neurons>
seq_lin: !new:speechbrain.nnet.linear.Linear
input_size: !ref <d_model>
n_neurons: !ref <output_neurons>
modules:
CNN: !ref <CNN>
Transformer: !ref <Transformer>
seq_lin: !ref <seq_lin>
ctc_lin: !ref <ctc_lin>
model: !new:torch.nn.ModuleList
- [!ref <CNN>, !ref <Transformer>, !ref <seq_lin>, !ref <ctc_lin>]
# define two optimizers here for two-stage training
Adam: !name:torch.optim.Adam
lr: 0
betas: (0.9, 0.98)
eps: 0.000000001
SGD: !name:torch.optim.SGD
lr: !ref <lr_sgd>
momentum: 0.99
nesterov: True
valid_search: !new:speechbrain.decoders.S2STransformerBeamSearch
modules: [!ref <Transformer>, !ref <seq_lin>, !ref <ctc_lin>]
bos_index: !ref <bos_index>
eos_index: !ref <eos_index>
blank_index: !ref <blank_index>
min_decode_ratio: !ref <min_decode_ratio>
max_decode_ratio: !ref <max_decode_ratio>
beam_size: !ref <valid_beam_size>
ctc_weight: !ref <ctc_weight_decode>
using_eos_threshold: False
length_normalization: False
test_search: !new:speechbrain.decoders.S2STransformerBeamSearch
modules: [!ref <Transformer>, !ref <seq_lin>, !ref <ctc_lin>]
bos_index: !ref <bos_index>
eos_index: !ref <eos_index>
blank_index: !ref <blank_index>
min_decode_ratio: !ref <min_decode_ratio>
max_decode_ratio: !ref <max_decode_ratio>
beam_size: !ref <test_beam_size>
ctc_weight: !ref <ctc_weight_decode>
lm_weight: !ref <lm_weight>
lm_modules: !ref <lm_model>
temperature: 1
temperature_lm: 1
using_eos_threshold: False
length_normalization: True
log_softmax: !new:torch.nn.LogSoftmax
dim: -1
ctc_cost: !name:speechbrain.nnet.losses.ctc_loss
blank_index: !ref <blank_index>
reduction: !ref <loss_reduction>
seq_cost: !name:speechbrain.nnet.losses.kldiv_loss
label_smoothing: !ref <label_smoothing>
reduction: !ref <loss_reduction>
noam_annealing: !new:speechbrain.nnet.schedulers.NoamScheduler
lr_initial: !ref <lr_adam>
n_warmup_steps: 25000
model_size: !ref <d_model>
checkpointer: !new:speechbrain.utils.checkpoints.Checkpointer
checkpoints_dir: !ref <save_folder>
recoverables:
model: !ref <model>
noam_scheduler: !ref <noam_annealing>
normalizer: !ref <normalize>
counter: !ref <epoch_counter>
epoch_counter: !new:speechbrain.utils.epoch_loop.EpochCounter
limit: !ref <number_of_epochs>
normalize: !new:speechbrain.processing.features.InputNormalization
norm_type: global
update_until_epoch: 4
augmentation: !new:speechbrain.lobes.augment.SpecAugment
time_warp: True
time_warp_window: 5
time_warp_mode: bicubic
freq_mask: True
n_freq_mask: 2
time_mask: True
n_time_mask: 2
replace_with_zero: False
freq_mask_width: 30
time_mask_width: 40
speed_perturb: !new:speechbrain.processing.speech_augmentation.SpeedPerturb
orig_freq: !ref <sample_rate>
speeds: [95, 100, 105]
compute_features: !new:speechbrain.lobes.features.Fbank
sample_rate: !ref <sample_rate>
n_fft: !ref <n_fft>
n_mels: !ref <n_mels>
train_logger: !new:speechbrain.utils.train_logger.FileTrainLogger
save_file: !ref <train_log>
error_rate_computer: !name:speechbrain.utils.metric_stats.ErrorRateStats
acc_computer: !name:speechbrain.utils.Accuracy.AccuracyStats
# The pretrainer allows a mapping between pretrained files and instances that
# are declared in the yaml. E.g here, we will download the file lm.ckpt
# and it will be loaded into "lm" which is pointing to the <lm_model> defined
# before.
pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer
collect_in: !ref <save_folder>
loadables:
lm: !ref <lm_model>
tokenizer: !ref <tokenizer>
paths:
lm: !ref <pretrained_lm_tokenizer_path>/lm.ckpt
tokenizer: !ref <pretrained_lm_tokenizer_path>/tokenizer.ckpt