flexthink
commited on
Commit
·
edde4dc
1
Parent(s):
8829ad6
Update to match the latest version of SpeechBrain
Browse files- hyperparams.yaml +23 -36
hyperparams.yaml
CHANGED
|
@@ -46,6 +46,7 @@ lexicon_batch_size: 1024
|
|
| 46 |
sentence_batch_size: 32
|
| 47 |
homograph_batch_size: 32
|
| 48 |
ctc_weight: 0.5
|
|
|
|
| 49 |
homograph_loss_weight: 2.0
|
| 50 |
lr: 0.002
|
| 51 |
save_for_pretrained: true
|
|
@@ -97,7 +98,6 @@ lm_layers: 2 # number of hidden layers
|
|
| 97 |
lm_output_neurons: 43
|
| 98 |
|
| 99 |
# Beam Searcher
|
| 100 |
-
use_language_model: false
|
| 101 |
beam_search_min_decode_ratio: 0
|
| 102 |
beam_search_max_decode_ratio: 1.0
|
| 103 |
beam_search_beam_size: 16
|
|
@@ -268,7 +268,7 @@ modules:
|
|
| 268 |
lin: *id010
|
| 269 |
ctc_lin: *id013
|
| 270 |
out: *id011
|
| 271 |
-
word_emb:
|
| 272 |
word_emb_enc: *id012
|
| 273 |
model: *id014
|
| 274 |
lm_model: &id015 !new:speechbrain.lobes.models.RNNLM.RNNLM
|
|
@@ -281,61 +281,48 @@ lm_model: &id015 !new:speechbrain.lobes.models.RNNLM.RNNLM
|
|
| 281 |
opt_class: !name:torch.optim.Adam
|
| 282 |
lr: 0.002
|
| 283 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 284 |
beam_searcher: &id029 !new:speechbrain.decoders.S2SRNNBeamSearcher
|
| 285 |
embedding: *id008
|
| 286 |
decoder: *id009
|
| 287 |
linear: *id010
|
| 288 |
-
|
| 289 |
-
|
| 290 |
-
eos_index: 1
|
| 291 |
-
blank_index: 2
|
| 292 |
min_decode_ratio: 0
|
| 293 |
max_decode_ratio: 1.0
|
| 294 |
beam_size: 16
|
| 295 |
eos_threshold: 10.0
|
| 296 |
using_max_attn_shift: false
|
| 297 |
max_attn_shift: 10
|
| 298 |
-
|
| 299 |
-
ctc_weight: 0.4
|
| 300 |
|
| 301 |
beam_searcher_valid: !new:speechbrain.decoders.S2SRNNBeamSearcher
|
| 302 |
embedding: *id008
|
| 303 |
decoder: *id009
|
| 304 |
linear: *id010
|
| 305 |
-
|
| 306 |
-
|
| 307 |
-
eos_index: 1
|
| 308 |
-
blank_index: 2
|
| 309 |
min_decode_ratio: 0
|
| 310 |
max_decode_ratio: 1.0
|
| 311 |
beam_size: 16
|
| 312 |
eos_threshold: 10.0
|
| 313 |
using_max_attn_shift: false
|
| 314 |
max_attn_shift: 10
|
| 315 |
-
|
| 316 |
-
ctc_weight: 0.4
|
| 317 |
-
|
| 318 |
-
beam_searcher_lm: !new:speechbrain.decoders.seq2seq.S2SRNNBeamSearchLM
|
| 319 |
-
embedding: *id008
|
| 320 |
-
decoder: *id009
|
| 321 |
-
linear: *id010
|
| 322 |
-
ctc_linear: *id013
|
| 323 |
-
language_model: *id015
|
| 324 |
-
bos_index: 0
|
| 325 |
-
eos_index: 1
|
| 326 |
-
blank_index: 2
|
| 327 |
-
min_decode_ratio: 0
|
| 328 |
-
max_decode_ratio: 1.0
|
| 329 |
-
beam_size: 16
|
| 330 |
-
eos_threshold: 10.0
|
| 331 |
-
using_max_attn_shift: false
|
| 332 |
-
max_attn_shift: 10
|
| 333 |
-
coverage_penalty: 5.0
|
| 334 |
-
ctc_weight: 0.4
|
| 335 |
-
lm_weight: 0.5
|
| 336 |
-
temperature: 1.25
|
| 337 |
-
temperature_lm: 1.0
|
| 338 |
-
|
| 339 |
|
| 340 |
lr_annealing: &id018 !new:speechbrain.nnet.schedulers.NewBobScheduler
|
| 341 |
initial_value: 0.002
|
|
|
|
| 46 |
sentence_batch_size: 32
|
| 47 |
homograph_batch_size: 32
|
| 48 |
ctc_weight: 0.5
|
| 49 |
+
ctc_window_size: 0
|
| 50 |
homograph_loss_weight: 2.0
|
| 51 |
lr: 0.002
|
| 52 |
save_for_pretrained: true
|
|
|
|
| 98 |
lm_output_neurons: 43
|
| 99 |
|
| 100 |
# Beam Searcher
|
|
|
|
| 101 |
beam_search_min_decode_ratio: 0
|
| 102 |
beam_search_max_decode_ratio: 1.0
|
| 103 |
beam_search_beam_size: 16
|
|
|
|
| 268 |
lin: *id010
|
| 269 |
ctc_lin: *id013
|
| 270 |
out: *id011
|
| 271 |
+
word_emb: !ref <word_emb>
|
| 272 |
word_emb_enc: *id012
|
| 273 |
model: *id014
|
| 274 |
lm_model: &id015 !new:speechbrain.lobes.models.RNNLM.RNNLM
|
|
|
|
| 281 |
opt_class: !name:torch.optim.Adam
|
| 282 |
lr: 0.002
|
| 283 |
|
| 284 |
+
ctc_scorer: !new:speechbrain.decoders.scorer.CTCScorer
|
| 285 |
+
eos_index: !ref <eos_index>
|
| 286 |
+
blank_index: !ref <blank_index>
|
| 287 |
+
ctc_fc: !ref <ctc_lin>
|
| 288 |
+
ctc_window_size: !ref <ctc_window_size>
|
| 289 |
+
|
| 290 |
+
coverage_scorer: !new:speechbrain.decoders.scorer.CoverageScorer
|
| 291 |
+
vocab_size: !ref <output_neurons>
|
| 292 |
+
|
| 293 |
+
scorer: !new:speechbrain.decoders.scorer.ScorerBuilder
|
| 294 |
+
full_scorers: [!ref <coverage_scorer>, !ref <ctc_scorer>]
|
| 295 |
+
weights:
|
| 296 |
+
coverage: !ref <beam_search_coverage_penalty>
|
| 297 |
+
ctc: !ref <ctc_weight>
|
| 298 |
+
|
| 299 |
beam_searcher: &id029 !new:speechbrain.decoders.S2SRNNBeamSearcher
|
| 300 |
embedding: *id008
|
| 301 |
decoder: *id009
|
| 302 |
linear: *id010
|
| 303 |
+
bos_index: !ref <bos_index>
|
| 304 |
+
eos_index: !ref <eos_index>
|
|
|
|
|
|
|
| 305 |
min_decode_ratio: 0
|
| 306 |
max_decode_ratio: 1.0
|
| 307 |
beam_size: 16
|
| 308 |
eos_threshold: 10.0
|
| 309 |
using_max_attn_shift: false
|
| 310 |
max_attn_shift: 10
|
| 311 |
+
scorer: !ref <scorer>
|
|
|
|
| 312 |
|
| 313 |
beam_searcher_valid: !new:speechbrain.decoders.S2SRNNBeamSearcher
|
| 314 |
embedding: *id008
|
| 315 |
decoder: *id009
|
| 316 |
linear: *id010
|
| 317 |
+
bos_index: !ref <bos_index>
|
| 318 |
+
eos_index: !ref <eos_index>
|
|
|
|
|
|
|
| 319 |
min_decode_ratio: 0
|
| 320 |
max_decode_ratio: 1.0
|
| 321 |
beam_size: 16
|
| 322 |
eos_threshold: 10.0
|
| 323 |
using_max_attn_shift: false
|
| 324 |
max_attn_shift: 10
|
| 325 |
+
scorer: !ref <scorer>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 326 |
|
| 327 |
lr_annealing: &id018 !new:speechbrain.nnet.schedulers.NewBobScheduler
|
| 328 |
initial_value: 0.002
|