-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathamharic_16000.yaml
executable file
·205 lines (185 loc) · 5.55 KB
/
amharic_16000.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
name: &name "QuartzNet15x5"
sample_rate: &sample_rate 16000
repeat: &repeat 1
dropout: &dropout 0.0
separable: &separable true
labels: &labels [' ', ሀ, ሁ, ሂ, ሃ, ሄ, ህ, ሆ, ለ, ሉ,
ሊ, ላ, ሌ, ል, ሎ, ሏ, ሐ, ሑ, ሒ, ሔ, ሕ, መ, ሙ,
ሚ, ማ, ሜ, ም, ሞ, ሟ, ሠ, ሡ, ሣ, ሥ, ሦ, ረ, ሩ,
ሪ, ራ, ሬ, ር, ሮ, ሯ, ሰ, ሱ, ሲ, ሳ, ሴ, ስ, ሶ,
ሷ, ሸ, ሹ, ሺ, ሻ, ሼ, ሽ, ሾ, ሿ, ቀ, ቁ, ቂ, ቃ,
ቄ, ቅ, ቆ, ቋ, በ, ቡ, ቢ, ባ, ቤ, ብ, ቦ, ቧ, ቨ,
ቪ, ቫ, ቬ, ቭ, ተ, ቱ, ቲ, ታ, ቴ, ት, ቶ, ቷ, ቸ,
ቹ, ቺ, ቻ, ቼ, ች, ቾ, ቿ, ኋ, ነ, ኑ, ኒ, ና, ኔ,
ን, ኖ, ኗ, ኘ, ኙ, ኚ, ኛ, ኜ, ኝ, ኞ, አ, ኡ, ኢ,
ኤ, እ, ኦ, ኧ, ከ, ኩ, ኪ, ካ, ኬ, ክ, ኮ, ኳ, ኸ,
ኻ, ኼ, ኽ, ወ, ዉ, ዊ, ዋ, ዌ, ው, ዎ, ዐ, ዑ, ዓ,
ዕ, ዘ, ዙ, ዚ, ዛ, ዜ, ዝ, ዞ, ዟ, ዠ, ዢ, ዣ, ዤ,
ዥ, ዦ, የ, ዩ, ዪ, ያ, ዬ, ይ, ዮ, ደ, ዱ, ዲ, ዳ,
ዴ, ድ, ዶ, ዷ, ጀ, ጁ, ጂ, ጃ, ጄ, ጅ, ጆ, ጇ, ገ,
ጉ, ጊ, ጋ, ጌ, ግ, ጎ, ጓ, ጠ, ጡ, ጢ, ጣ, ጤ, ጥ,
ጦ, ጧ, ጨ, ጩ, ጪ, ጫ, ጬ, ጭ, ጮ, ጱ, ጲ, ጳ, ጴ,
ጵ, ጸ, ጹ, ጺ, ጻ, ጼ, ጽ, ጾ, ጿ, ፀ, ፁ, ፃ, ፄ,
ፅ, ፆ, ፈ, ፉ, ፊ, ፋ, ፌ, ፍ, ፎ, ፏ, ፐ, ፑ, ፒ,
ፓ, ፔ, ፕ, ፖ]
model:
train_ds:
manifest_filepath: ???
sample_rate: 16000
labels: *labels
batch_size: 12
trim_silence: True
max_duration: 33.0
shuffle: True
is_tarred: False
tarred_audio_filepaths: null
num_workers: 16
validation_ds:
manifest_filepath: ???
sample_rate: 16000
labels: *labels
batch_size: 12
shuffle: False
num_workers: 16
preprocessor:
cls: nemo.collections.asr.modules.AudioToMelSpectrogramPreprocessor
params:
normalize: "per_feature"
window_size: 0.02
sample_rate: *sample_rate
window_stride: 0.01
window: "hann"
features: &n_mels 64
n_fft: 512
frame_splicing: 1
dither: 0.00001
stft_conv: false
spec_augment:
cls: nemo.collections.asr.modules.SpectrogramAugmentation
params:
rect_freq: 50
rect_masks: 5
rect_time: 120
encoder:
cls: nemo.collections.asr.modules.ConvASREncoder
params:
feat_in: *n_mels
activation: relu
conv_mask: true
jasper:
- filters: 128
repeat: 1
kernel: [11]
stride: [1]
dilation: [1]
dropout: *dropout
residual: true
separable: *separable
se: true
se_context_size: -1
- filters: 256
repeat: *repeat
kernel: [13]
stride: [1]
dilation: [1]
dropout: *dropout
residual: true
separable: *separable
se: true
se_context_size: -1
- filters: 256
repeat: *repeat
kernel: [15]
stride: [1]
dilation: [1]
dropout: *dropout
residual: true
separable: *separable
se: true
se_context_size: -1
- filters: 256
repeat: *repeat
kernel: [17]
stride: [1]
dilation: [1]
dropout: *dropout
residual: true
separable: *separable
se: true
se_context_size: -1
- filters: 256
repeat: *repeat
kernel: [19]
stride: [1]
dilation: [1]
dropout: *dropout
residual: true
separable: *separable
se: true
se_context_size: -1
- filters: 256
repeat: 1
kernel: [21]
stride: [1]
dilation: [1]
dropout: 0.0
residual: false
separable: *separable
se: true
se_context_size: -1
- filters: &enc_feat_out 1024
repeat: 1
kernel: [1]
stride: [1]
dilation: [1]
dropout: 0.0
residual: false
separable: *separable
se: true
se_context_size: -1
decoder:
cls: nemo.collections.asr.modules.ConvASRDecoder
params:
feat_in: 1024
num_classes: 28
vocabulary: *labels
optim:
name: novograd
# cls: nemo.core.optim.optimizers.Novograd
lr: .00005
# optimizer arguments
betas: [0.8, 0.5]
weight_decay: 0.001
# scheduler setup
sched:
name: CosineAnnealing
# pytorch lightning args
monitor: val_loss
reduce_on_plateau: false
# Scheduler params
warmup_steps: null
warmup_ratio: null
min_lr: 0.0
last_epoch: -1
trainer:
gpus: 0 # number of gpus
max_epochs: 5
max_steps: null # computed at runtime if not set
num_nodes: 1
distributed_backend: ddp
accumulate_grad_batches: 1
checkpoint_callback: False # Provided by exp_manager
logger: False # Provided by exp_manager
row_log_interval: 100 # Interval of logging.
val_check_interval: 1.0 # check once per epoch .25 for 4 times per epoch
exp_manager:
exp_dir: null
name: *name
create_tensorboard_logger: True
create_checkpoint_callback: True
hydra:
run:
dir: .
job_logging:
root:
handlers: null