File size: 11,324 Bytes
0ad8d41
 
 
 
 
 
 
 
 
 
 
 
6297333
0ad8d41
 
 
6297333
0ad8d41
 
6297333
 
0ad8d41
 
6297333
0ad8d41
 
 
 
 
6297333
 
0ad8d41
edde4dc
0ad8d41
 
 
6297333
 
0ad8d41
6297333
 
0ad8d41
6297333
 
0ad8d41
6297333
 
0ad8d41
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6297333
0ad8d41
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6297333
0ad8d41
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6297333
 
 
 
0ad8d41
6297333
 
0ad8d41
6297333
 
0ad8d41
6297333
 
 
 
 
 
0ad8d41
6297333
 
 
0ad8d41
 
6297333
 
 
 
 
 
 
0ad8d41
6297333
 
 
0ad8d41
6297333
 
 
0ad8d41
 
6297333
 
 
 
 
 
 
 
0ad8d41
 
 
 
 
 
6297333
0ad8d41
 
6297333
 
 
 
 
 
 
 
 
 
0ad8d41
6297333
 
 
 
 
 
 
 
edde4dc
6297333
 
 
 
 
 
 
 
0ad8d41
edde4dc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6297333
 
 
 
edde4dc
 
6297333
 
 
 
 
 
 
edde4dc
0ad8d41
 
6297333
 
 
edde4dc
 
6297333
 
 
 
 
 
 
edde4dc
0ad8d41
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6297333
 
 
 
 
0ad8d41
6297333
 
0ad8d41
 
 
6297333
0ad8d41
 
6297333
 
 
 
 
0ad8d41
6297333
 
0ad8d41
6297333
0ad8d41
6297333
 
 
 
 
0ad8d41
6297333
 
0ad8d41
 
 
 
6297333
 
0ad8d41
 
 
 
 
 
 
 
 
 
6297333
 
 
 
0ad8d41
6297333
 
 
 
0ad8d41
 
6297333
0ad8d41
 
 
6297333
0ad8d41
 
 
 
 
 
 
6297333
0ad8d41
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6297333
0ad8d41
 
 
 
 
 
 
 
 
6297333
 
 
 
 
 
 
0ad8d41
 
 
 
 
 
 
 
6297333
 
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
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
# Generated 2022-07-09 from:
# /notebooks/speechbrain/recipes/LibriSpeech/G2P/hparams/hparams_g2p_rnn.yaml
# yamllint disable
# ################################
# Model: LSTM (encoder) + GRU (decoder) (tokenized)
# Authors:
# Loren Lugosch & Mirco Ravanelli 2020
# Artem Ploujnikov 2021
# ################################

# Seed needs to be set at top of yaml, before objects with parameters are made
seed: 1234
__set_seed: !apply:torch.manual_seed [!ref <seed>]


# Tokenizers
char_tokenize: False
char_token_type: unigram  # ["unigram", "bpe", "char"]
char_token_output: 512
char_token_wordwise: True
phn_tokenize: False
phn_token_type: unigram  # ["unigram", "bpe", "char"]
phn_token_output: 512  # index(blank/eos/bos/unk) = 0
phn_token_wordwise: True
character_coverage: 1.0


phonemes_count: 43
graphemes_count: 31
phonemes_enable_space: True

ctc_weight: 0.5
ctc_window_size: 0
homograph_loss_weight: 2.0

# Model parameters
output_neurons: !apply:speechbrain.utils.hparams.choice
  value: !ref <phn_tokenize>
  choices:
    True: !ref <phn_token_output> + 1
    False: !ref <phonemes_count>

enc_num_embeddings: !apply:speechbrain.utils.hparams.choice
  value: !ref <char_tokenize>
  choices:
    True: !ref <char_token_output> + 1
    False: !ref <graphemes_count>

enc_dropout: 0.5
enc_neurons: 512
enc_num_layers: 4
dec_dropout: 0.5
dec_neurons: 512
dec_att_neurons: 256
dec_num_layers: 4
embedding_dim: 512

# Determines whether to use BOS (beginning-of-sequence) or EOS (end-of-sequence) tokens
# Available modes:
# raw: no BOS/EOS tokens are added
# bos: a beginning-of-sequence token is added
# eos: an end-of-sequence token is added
grapheme_sequence_mode: bos
phoneme_sequence_mode: bos


# Special Token information
bos_index: 0
eos_index: 1
blank_index: 2
unk_index: 2
token_space_index: 512


# Language Model
lm_emb_dim: 256 # dimension of the embeddings
lm_rnn_size: 512 # dimension of hidden layers
lm_layers: 2 # number of hidden layers
lm_output_neurons: 43

# Beam Searcher
beam_search_min_decode_ratio: 0
beam_search_max_decode_ratio: 1.0
beam_search_beam_size: 16
beam_search_beam_size_valid: 16
beam_search_eos_threshold: 10.0
beam_search_using_max_attn_shift: false
beam_search_max_attn_shift: 10
beam_search_coverage_penalty: 5.0
beam_search_lm_weight: 0.5
beam_search_ctc_weight_decode: 0.4
beam_search_temperature: 1.25
beam_search_temperature_lm: 1.0

# Word embeddings
use_word_emb: true
word_emb_model: bert-base-uncased
word_emb_dim: 768
word_emb_enc_dim: 256
word_emb_norm_type: batch

graphemes:
- A
- B
- C
- D
- E
- F
- G
- H
- I
- J
- K
- L
- M
- N
- O
- P
- Q
- R
- S
- T
- U
- V
- W
- X
- Y
- Z
- "'"
- ' '

phonemes: 
- AA
- AE
- AH
- AO
- AW
- AY
- B
- CH
- D
- DH
- EH
- ER
- EY
- F
- G
- HH
- IH
- IY
- JH
- K
- L
- M
- N
- NG
- OW
- OY
- P
- R
- S
- SH
- T
- TH
- UH
- UW
- V
- W
- Y
- Z
- ZH
- ' '

enc_input_dim: !apply:speechbrain.lobes.models.g2p.model.input_dim
  use_word_emb: !ref <use_word_emb>
  word_emb_enc_dim: !ref <word_emb_enc_dim>
  embedding_dim: !ref <embedding_dim>

phn_char_map: !apply:speechbrain.lobes.models.g2p.dataio.build_token_char_map
  tokens: !ref <phonemes>

char_phn_map: !apply:speechbrain.lobes.models.g2p.dataio.flip_map
  map_dict: !ref <phn_char_map>

enc: !new:speechbrain.nnet.RNN.LSTM
  input_shape: [null, null, !ref <enc_input_dim>]
  bidirectional: True
  hidden_size: !ref <enc_neurons>
  num_layers: !ref <enc_num_layers>
  dropout: !ref <enc_dropout>

lin: !new:speechbrain.nnet.linear.Linear
  input_size: !ref <dec_neurons>
  n_neurons: !ref <output_neurons>
  bias: false

ctc_lin: !new:speechbrain.nnet.linear.Linear
  input_size: !ref 2 * <enc_neurons>
  n_neurons: !ref <output_neurons>

encoder_emb: !new:speechbrain.nnet.embedding.Embedding
  num_embeddings: !ref <enc_num_embeddings>
  embedding_dim: !ref <embedding_dim>

emb: !new:speechbrain.nnet.embedding.Embedding
  num_embeddings: !ref <output_neurons>
  embedding_dim: !ref <embedding_dim>

dec: !new:speechbrain.nnet.RNN.AttentionalRNNDecoder
  enc_dim: !ref <enc_neurons> * 2
  input_size: !ref <embedding_dim>
  rnn_type: gru
  attn_type: content
  dropout: !ref <dec_dropout>
  hidden_size: !ref <dec_neurons>
  attn_dim: !ref <dec_att_neurons>
  num_layers: !ref <dec_num_layers>

word_emb_enc: !new:speechbrain.lobes.models.g2p.model.WordEmbeddingEncoder
  word_emb_dim: !ref <word_emb_dim>
  word_emb_enc_dim: !ref <word_emb_enc_dim>
  norm_type: batch

word_emb: !apply:speechbrain.lobes.models.g2p.dataio.lazy_init
  init: !name:speechbrain.wordemb.transformer.TransformerWordEmbeddings
    model: bert-base-uncased

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

model: !new:speechbrain.lobes.models.g2p.model.AttentionSeq2Seq
  enc: !ref <enc>
  encoder_emb: !ref <encoder_emb>
  emb: !ref <emb>
  dec: !ref <dec>
  lin: !ref <lin>
  out: !ref <log_softmax>
  use_word_emb: !ref <use_word_emb>
  word_emb_enc: !ref <word_emb_enc>

modules:
  model: !ref <model>
  enc: !ref <enc>
  encoder_emb: !ref <encoder_emb>
  emb: !ref <emb>
  dec: !ref <dec>
  lin: !ref <lin>
  ctc_lin: !ref <ctc_lin>
  out: !ref <log_softmax>
  word_emb: !ref <word_emb>
  word_emb_enc: !ref <word_emb_enc>

lm_model: !new:speechbrain.lobes.models.RNNLM.RNNLM
  embedding_dim: !ref <lm_emb_dim>
  rnn_layers: !ref <lm_layers>
  rnn_neurons: !ref <lm_rnn_size>
  output_neurons: !ref <lm_output_neurons>
  return_hidden: True

ctc_scorer: !new:speechbrain.decoders.scorer.CTCScorer
  eos_index: !ref <eos_index>
  blank_index: !ref <blank_index>
  ctc_fc: !ref <ctc_lin>
  ctc_window_size: !ref <ctc_window_size>

coverage_scorer: !new:speechbrain.decoders.scorer.CoverageScorer
   vocab_size: !ref <output_neurons>

scorer: !new:speechbrain.decoders.scorer.ScorerBuilder
   full_scorers: [!ref <coverage_scorer>, !ref <ctc_scorer>]
   weights:
      coverage: !ref <beam_search_coverage_penalty>
      ctc: !ref <ctc_weight>

beam_searcher: !new:speechbrain.decoders.S2SRNNBeamSearcher
  embedding: !ref <emb>
  decoder: !ref <dec>
  linear: !ref <lin>
  bos_index: !ref <bos_index>
  eos_index: !ref <eos_index>
  min_decode_ratio: !ref <beam_search_min_decode_ratio>
  max_decode_ratio: !ref <beam_search_max_decode_ratio>
  beam_size: !ref <beam_search_beam_size>
  eos_threshold: !ref <beam_search_eos_threshold>
  using_max_attn_shift: !ref <beam_search_using_max_attn_shift>
  max_attn_shift: !ref <beam_search_max_attn_shift>
  temperature: !ref <beam_search_temperature>
  scorer: !ref <scorer>

beam_searcher_valid: !new:speechbrain.decoders.S2SRNNBeamSearcher
  embedding: !ref <emb>
  decoder: !ref <dec>
  linear: !ref <lin>
  bos_index: !ref <bos_index>
  eos_index: !ref <eos_index>
  min_decode_ratio: !ref <beam_search_min_decode_ratio>
  max_decode_ratio: !ref <beam_search_max_decode_ratio>
  beam_size: !ref <beam_search_beam_size>
  eos_threshold: !ref <beam_search_eos_threshold>
  using_max_attn_shift: !ref <beam_search_using_max_attn_shift>
  max_attn_shift: !ref <beam_search_max_attn_shift>
  temperature: !ref <beam_search_temperature>
  scorer: !ref <scorer>

homograph_extractor: !new:speechbrain.lobes.models.g2p.homograph.SubsequenceExtractor

model_output_keys:
- p_seq
- char_lens
- encoder_out

grapheme_encoder: &id027 !new:speechbrain.dataio.encoder.TextEncoder
phoneme_encoder: &id024 !new:speechbrain.dataio.encoder.TextEncoder


grapheme_tokenizer: !apply:speechbrain.lobes.models.g2p.dataio.lazy_init
  init: !name:speechbrain.tokenizers.SentencePiece.SentencePiece
    model_dir: grapheme_tokenizer
    bos_id: !ref <bos_index>
    eos_id: !ref <eos_index>
    unk_id: !ref <unk_index>
    vocab_size: !ref <char_token_output>
    annotation_train: null
    annotation_read: char
    model_type: !ref <char_token_type> # ["unigram", "bpe", "char"]
    character_coverage: !ref <character_coverage>
    annotation_format: json
    text_file: grapheme_annotations.txt

phoneme_tokenizer: !apply:speechbrain.lobes.models.g2p.dataio.lazy_init
  init: !name:speechbrain.tokenizers.SentencePiece.SentencePiece
    model_dir: phoneme_tokenizer
    bos_id: !ref <bos_index>
    eos_id: !ref <eos_index>
    unk_id: !ref <unk_index>
    vocab_size: !ref <phn_token_output>
    annotation_train: null
    annotation_read: phn
    model_type: !ref <phn_token_type> # ["unigram", "bpe", "char"]
    character_coverage: !ref <character_coverage>
    annotation_format: json
    text_file: null

out_phoneme_decoder_tok: !apply:speechbrain.lobes.models.g2p.dataio.char_map_detokenize
  tokenizer: !ref <phoneme_tokenizer>
  char_map: !ref <char_phn_map>
  token_space_index: !ref <token_space_index>
  wordwise: !ref <phn_token_wordwise>

out_phoneme_decoder_raw:  !name:speechbrain.lobes.models.g2p.dataio.text_decode
  encoder: !ref <phoneme_encoder>

out_phoneme_decoder: !apply:speechbrain.utils.hparams.choice
  value: false
  choices:
    True: !ref <out_phoneme_decoder_tok>
    False: !ref <out_phoneme_decoder_raw>
encode_pipeline:
  batch: false
  use_padded_data: true
  output_keys:
  - grapheme_list
  - grapheme_encoded_list
  - grapheme_encoded
  - word_emb
  init:
  - func: !name:speechbrain.lobes.models.g2p.dataio.enable_eos_bos
      encoder: !ref <grapheme_encoder>
      tokens: !ref <graphemes>
      bos_index: !ref <bos_index>
      eos_index: !ref <eos_index>
  - func: !name:speechbrain.lobes.models.g2p.dataio.enable_eos_bos
      encoder: !ref <phoneme_encoder>
      tokens: !ref <phonemes>
      bos_index: !ref <bos_index>
      eos_index: !ref <eos_index>
  steps:
  - func: !name:speechbrain.lobes.models.g2p.dataio.clean_pipeline
      graphemes: !ref <graphemes>
    takes: txt
    provides: txt_cleaned
  - func: !name:speechbrain.lobes.models.g2p.dataio.grapheme_pipeline
      grapheme_encoder: !ref <grapheme_encoder>
    takes: txt_cleaned
    provides:
    - grapheme_list
    - grapheme_encoded_list
    - grapheme_encoded_raw

  - func: !name:speechbrain.lobes.models.g2p.dataio.add_bos_eos
      encoder: !ref <grapheme_encoder>
    takes: grapheme_encoded_list
    provides:
    - grapheme_encoded
    - grapheme_len
    - grapheme_encoded_eos
    - grapheme_len_eos
  - func: !name:speechbrain.lobes.models.g2p.dataio.word_emb_pipeline
      word_emb: !ref <word_emb>
      grapheme_encoder: !ref <grapheme_encoder>
      use_word_emb: !ref <use_word_emb>
    takes:
    - txt
    - grapheme_encoded
    - grapheme_len
    provides: word_emb

decode_pipeline:
  batch: true
  output_keys:
  - phonemes
  steps:
  - func: !name:speechbrain.lobes.models.g2p.dataio.beam_search_pipeline
      beam_searcher: !ref <beam_searcher>
    takes:
    - char_lens
    - encoder_out
    provides:
    - hyps
    - scores
  - func: !apply:speechbrain.utils.hparams.choice
      value: false
      choices:
        True: !apply:speechbrain.lobes.models.g2p.dataio.char_map_detokenize
          tokenizer: !ref <phoneme_tokenizer>
          char_map: !ref <char_phn_map>
          token_space_index: !ref <token_space_index>
          wordwise: !ref <phn_token_wordwise>
        False: !name:speechbrain.lobes.models.g2p.dataio.phoneme_decoder_pipeline
          phoneme_encoder: !ref <phoneme_encoder>
    takes:
    - hyps
    provides:
    - phonemes


pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer
  loadables:
    model: !ref <model>
    ctc_lin: !ref <ctc_lin>