File size: 5,393 Bytes
b2049a6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# Seed needs to be set at top of yaml, before objects with parameters are made
seed: 1987
__set_seed: !apply:torch.manual_seed [!ref <seed>]

lang_csv: Swahili

output_folder: !ref results/finetune_hubert_ASR_char/<seed>/<lang_csv>
output_wer_folder: !ref <output_folder>/
save_folder: !ref <output_folder>/save
train_log: !ref <output_folder>/train_log.txt

# huggingface format 
hubert_hub: Orange/SSA-HuBERT-base-5k

hubert_folder: !ref <save_folder>/hubert_checkpoint

# Data files
data_folder: !ref PATH_TO_YOUR_FOLDER/data_speechbrain/<lang_csv>

ckpt_interval_minutes: 10 # save checkpoint every N min
train_csv: !ref <data_folder>/train.csv
valid_csv: !ref <data_folder>/validation.csv 
test_csv:
        - !ref <data_folder>/test.csv

####################### Training Parameters ####################################

number_of_epochs: 10
lr: 0.1
lr_hubert: 0.000005
sorting: ascending
precision: fp32 # bf16, fp16 or fp32
sample_rate: 16000

# skip audio file longer than
avoid_if_longer_than: 60

batch_size: 2
test_batch_size: 2

# Dataloader options
train_dataloader_opts:
   batch_size: !ref <batch_size>

valid_dataloader_opts:
   batch_size: !ref <batch_size>

test_dataloader_opts:
   batch_size: !ref <test_batch_size>

####################### Model Parameters #######################################
activation: !name:torch.nn.LeakyReLU
dnn_layers: 2
dnn_neurons: 1024
freeze_hubert: False

# Outputs
output_neurons: 66  # BPE size, index(blank/eos/bos) = 0
blank_index: 0

#
# Functions and classes
#

label_encoder: !new:speechbrain.dataio.encoder.CTCTextEncoder

epoch_counter: !new:speechbrain.utils.epoch_loop.EpochCounter
   limit: !ref <number_of_epochs>

hubert: !new:speechbrain.lobes.models.huggingface_transformers.hubert.HuBERT
   source: !ref <hubert_hub>
   output_norm: True
   freeze: !ref <freeze_hubert>
   save_path: !ref <hubert_folder>
   
top_lin: !new:speechbrain.lobes.models.VanillaNN.VanillaNN
   input_shape: [null, null, 768] # 768 == output of hubert base model
   activation: !ref <activation>
   dnn_blocks: !ref <dnn_layers>
   dnn_neurons: !ref <dnn_neurons>

ctc_lin: !new:speechbrain.nnet.linear.Linear
   input_size: !ref <dnn_neurons>
   n_neurons: !ref <output_neurons>

log_softmax: !new:speechbrain.nnet.activations.Softmax
   apply_log: True

ctc_cost: !name:speechbrain.nnet.losses.ctc_loss
   blank_index: !ref <blank_index>

modules:
   hubert: !ref <hubert>
   top_lin: !ref <top_lin>
   ctc_lin: !ref <ctc_lin>

model: !new:torch.nn.ModuleList
   - [!ref <top_lin>, !ref <ctc_lin>]

model_opt_class: !name:torch.optim.Adadelta
   lr: !ref <lr>
   rho: 0.95
   eps: 1.e-8

hubert_opt_class: !name:torch.optim.Adam
   lr: !ref <lr_hubert>

lr_annealing_model: !new:speechbrain.nnet.schedulers.NewBobScheduler
   initial_value: !ref <lr>
   improvement_threshold: 0.0025
   annealing_factor: 0.8
   patient: 0

lr_annealing_hubert: !new:speechbrain.nnet.schedulers.NewBobScheduler
   initial_value: !ref <lr_hubert>
   improvement_threshold: 0.0025
   annealing_factor: 0.9
   patient: 0

############################## Augmentations ###################################

# 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 <speed_perturb>,
      !ref <drop_freq>,
      !ref <drop_chunk>]

############################## Decoding ########################################

# Decoding parameters
test_beam_search:
   beam_size: 143
   topk: 1
   blank_index: !ref <blank_index>
   space_token: ' ' # make sure this is the same as the one used in the tokenizer
   beam_prune_logp: -12.0
   token_prune_min_logp: -1.20
   prune_history: True
   alpha: 0.8
   beta: 1.2
   # can be downloaded from here https://www.openslr.org/11/ or trained with kenLM
   # It can either be a .bin or .arpa ; note: .arpa is much slower at loading
   # If you don't want to use an LM, comment it out or set it to null
   kenlm_model_path: null

############################## Logging and Pretrainer ##########################

checkpointer: !new:speechbrain.utils.checkpoints.Checkpointer
   checkpoints_dir: !ref <save_folder>
   recoverables:
      hubert: !ref <hubert>
      model: !ref <model>
      scheduler_model: !ref <lr_annealing_model>
      scheduler_hubert: !ref <lr_annealing_hubert>
      counter: !ref <epoch_counter>
      tokenizer: !ref <label_encoder>

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