forked from speechbrain/speechbrain
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.yaml
220 lines (184 loc) · 7.05 KB
/
train.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
# ############################################################################
# Model: Direct SLU
# Encoder: Pre-trained ASR encoder -> LSTM
# Decoder: GRU + beamsearch
# Tokens: BPE with unigram
# losses: NLL
# Training: Timers and Such
# Authors: Loren Lugosch, Mirco Ravanelli 2020
# ############################################################################
# Seed needs to be set at top of yaml, before objects with parameters are made
seed: 1986
__set_seed: !apply:speechbrain.utils.seed_everything [!ref <seed>]
experiment: train-real-and-synth
output_folder: !ref results/<experiment>/<seed>
save_folder: !ref <output_folder>/save
train_log: !ref <output_folder>/train_log.txt
all_real_wer_file: !ref <output_folder>/all_real_wer.txt
test_real_wer_file: !ref <output_folder>/test_real_wer.txt
test_synth_wer_file: !ref <output_folder>/test_synth_wer.txt
# Data files
data_folder: !PLACEHOLDER # e.g, /localscratch/timers-and-such
train_splits: ["train-synth", "train-real"]
csv_train: !ref <save_folder>/train-type=direct.csv
csv_dev_real: !ref <save_folder>/dev-real-type=direct.csv
csv_dev_synth: !ref <save_folder>/dev-synth-type=direct.csv
csv_test_real: !ref <save_folder>/test-real-type=direct.csv
csv_test_synth: !ref <save_folder>/test-synth-type=direct.csv
csv_all_real: !ref <save_folder>/all-real-type=direct.csv
tokenizer_file: https://huggingface.co/speechbrain/slu-timers-and-such-direct-librispeech-asr/resolve/main/tokenizer.ckpt
skip_prep: False
ckpt_interval_minutes: 15 # save checkpoint every N min
test_on_all_real: False
# Data for augmentation
data_folder_noise: !ref <data_folder>/noise # The noisy sequences for data augmentation will automatically be downloaded here.
NOISE_DATASET_URL: https://www.dropbox.com/scl/fi/a09pj97s5ifan81dqhi4n/noises.zip?rlkey=j8b0n9kdjdr32o1f06t0cw5b7&dl=1
noise_annotation: !ref <save_folder>/noise.csv #The data manifest files are created by the data preparation script
####################### Training Parameters ####################################
number_of_epochs: 1
batch_size: 16
lr: 0.0003
# token_type: unigram # ["unigram", "bpe", "char"]
sorting: random
####################### Model Parameters #######################################
sample_rate: 16000
emb_size: 128
dec_neurons: 512
output_neurons: 51 # index(eos/bos) = 0
ASR_encoder_dim: 512
encoder_dim: 256
# Decoding parameters
bos_index: 0
eos_index: 0
min_decode_ratio: 0.0
max_decode_ratio: 10.0
slu_beam_size: 80
eos_threshold: 1.5
temperature: 1.25
num_workers: 4
dataloader_opts:
num_workers: !ref <num_workers>
batch_size: !ref <batch_size>
shuffle: True
epoch_counter: !new:speechbrain.utils.epoch_loop.EpochCounter
limit: !ref <number_of_epochs>
############################## Augmentations ###################################
# Download and prepare the dataset of noisy sequences for augmentation
prepare_noise_data: !name:speechbrain.augment.preparation.prepare_dataset_from_URL
URL: !ref <NOISE_DATASET_URL>
dest_folder: !ref <data_folder_noise>
ext: wav
csv_file: !ref <noise_annotation>
# Add noise to input signal
add_noise: !new:speechbrain.augment.time_domain.AddNoise
csv_file: !ref <noise_annotation>
snr_low: 0
snr_high: 15
noise_sample_rate: !ref <sample_rate>
clean_sample_rate: !ref <sample_rate>
num_workers: !ref <num_workers>
# Speed perturbation
speed_perturb: !new:speechbrain.augment.time_domain.SpeedPerturb
orig_freq: !ref <sample_rate>
speeds: [95, 100, 105]
# Frequency drop: randomly drops a number of frequency bands to zero.
drop_freq: !new:speechbrain.augment.time_domain.DropFreq
drop_freq_low: 0
drop_freq_high: 1
drop_freq_count_low: 1
drop_freq_count_high: 3
drop_freq_width: 0.05
# Time drop: randomly drops a number of temporal chunks.
drop_chunk: !new:speechbrain.augment.time_domain.DropChunk
drop_length_low: 1000
drop_length_high: 2000
drop_count_low: 1
drop_count_high: 5
# Augmenter: Combines previously defined augmentations to perform data augmentation
wav_augment: !new:speechbrain.augment.augmenter.Augmenter
concat_original: True
min_augmentations: 4
max_augmentations: 4
augment_prob: 1.0
augmentations: [
!ref <add_noise>,
!ref <speed_perturb>,
!ref <drop_freq>,
!ref <drop_chunk>]
############################## Models ##########################################
asr_model_source: speechbrain/asr-crdnn-rnnlm-librispeech
slu_enc: !new:speechbrain.nnet.containers.Sequential
input_shape: [null, null, !ref <ASR_encoder_dim>]
lstm: !new:speechbrain.nnet.RNN.LSTM
input_size: !ref <ASR_encoder_dim>
bidirectional: True
hidden_size: !ref <encoder_dim>
num_layers: 2
linear: !new:speechbrain.nnet.linear.Linear
input_size: !ref <encoder_dim> * 2
n_neurons: !ref <encoder_dim>
output_emb: !new:speechbrain.nnet.embedding.Embedding
num_embeddings: !ref <output_neurons>
embedding_dim: !ref <emb_size>
dec: !new:speechbrain.nnet.RNN.AttentionalRNNDecoder
enc_dim: !ref <encoder_dim>
input_size: !ref <emb_size>
rnn_type: gru
attn_type: keyvalue
hidden_size: !ref <dec_neurons>
attn_dim: 512
num_layers: 3
scaling: 1.0
dropout: 0.0
seq_lin: !new:speechbrain.nnet.linear.Linear
input_size: !ref <dec_neurons>
n_neurons: !ref <output_neurons>
modules:
slu_enc: !ref <slu_enc>
output_emb: !ref <output_emb>
dec: !ref <dec>
seq_lin: !ref <seq_lin>
model: !new:torch.nn.ModuleList
- [!ref <slu_enc>, !ref <output_emb>,
!ref <dec>, !ref <seq_lin>]
tokenizer: !new:sentencepiece.SentencePieceProcessor
pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer
collect_in: !ref <save_folder>/TAS_tokenizer
loadables:
tokenizer: !ref <tokenizer>
paths:
tokenizer: !ref <tokenizer_file>
beam_searcher: !new:speechbrain.decoders.S2SRNNBeamSearcher
embedding: !ref <output_emb>
decoder: !ref <dec>
linear: !ref <seq_lin>
bos_index: !ref <bos_index>
eos_index: !ref <eos_index>
min_decode_ratio: !ref <min_decode_ratio>
max_decode_ratio: !ref <max_decode_ratio>
beam_size: !ref <slu_beam_size>
eos_threshold: !ref <eos_threshold>
temperature: !ref <temperature>
using_max_attn_shift: False
opt_class: !name:torch.optim.Adam
lr: !ref <lr>
lr_annealing: !new:speechbrain.nnet.schedulers.NewBobScheduler
initial_value: !ref <lr>
improvement_threshold: 0.0025
annealing_factor: 0.8
patient: 0
checkpointer: !new:speechbrain.utils.checkpoints.Checkpointer
checkpoints_dir: !ref <save_folder>
recoverables:
model: !ref <model>
scheduler: !ref <lr_annealing>
counter: !ref <epoch_counter>
log_softmax: !new:speechbrain.nnet.activations.Softmax
apply_log: True
seq_cost: !name:speechbrain.nnet.losses.nll_loss
label_smoothing: 0.1
train_logger: !new:speechbrain.utils.train_logger.FileTrainLogger
save_file: !ref <train_log>
error_rate_computer: !name:speechbrain.utils.metric_stats.ErrorRateStats
cer_computer: !name:speechbrain.utils.metric_stats.ErrorRateStats
split_tokens: True