|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
seed: 1986 |
|
__set_seed: !apply:torch.manual_seed [1986] |
|
output_folder: partly_frozen_splitted_wavlm |
|
wer_file: partly_frozen_splitted_wavlm/wer.txt |
|
save_folder: partly_frozen_splitted_wavlm/save |
|
train_log: partly_frozen_splitted_wavlm/train_log.txt |
|
|
|
|
|
|
|
|
|
data_folder: /gpfsscratch/rech/nou/uzn19yk/Libri/LibriSpeech/ |
|
|
|
data_folder_rirs: /gpfsscratch/rech/nou/uzn19yk/Libri/LibriSpeech/ |
|
train_splits: [train-clean-100] |
|
dev_splits: [dev-clean] |
|
test_splits: [test-clean, test-other] |
|
skip_prep: false |
|
ckpt_interval_minutes: 25 |
|
csv_folder: /gpfsstore/rech/nou/uzn19yk/iwslt/splitted_clean_tunisian_csvs/ |
|
train_csv: test_salah_local.csv |
|
valid_csv: test_salah_local.csv |
|
test_csv: |
|
- test_salah_local.csv |
|
|
|
|
|
number_of_epochs: 12 |
|
lr: 1 |
|
lr_wav2vec: 0.0001 |
|
sorting: ascending |
|
auto_mix_prec: false |
|
sample_rate: 16000 |
|
|
|
avoid_if_longer_than: 10 |
|
|
|
|
|
|
|
batch_size: 1 |
|
test_batch_size: 1 |
|
|
|
|
|
train_dataloader_opts: |
|
batch_size: 1 |
|
|
|
valid_dataloader_opts: |
|
batch_size: 1 |
|
|
|
test_dataloader_opts: |
|
batch_size: 1 |
|
|
|
|
|
activation: &id001 !name:torch.nn.LeakyReLU |
|
dnn_layers: 2 |
|
dnn_neurons: 1024 |
|
freeze_wav2vec: false |
|
|
|
|
|
output_neurons: 41 |
|
|
|
|
|
blank_index: 0 |
|
bos_index: 1 |
|
eos_index: 2 |
|
|
|
|
|
|
|
|
|
epoch_counter: &id008 !new:speechbrain.utils.epoch_loop.EpochCounter |
|
|
|
limit: 12 |
|
|
|
augmentation: !new:speechbrain.lobes.augment.TimeDomainSpecAugment |
|
sample_rate: 16000 |
|
speeds: [95, 100, 105] |
|
|
|
enc: &id003 !new:speechbrain.lobes.models.VanillaNN.VanillaNN |
|
input_shape: [null, null, 1024] |
|
activation: *id001 |
|
dnn_blocks: 2 |
|
dnn_neurons: 1024 |
|
|
|
wav2vec2: &id002 !new:speechbrain.lobes.models.huggingface_wav2vec.HuggingFaceWav2Vec2 |
|
source: wavlm-large/ |
|
output_norm: true |
|
freeze: false |
|
freeze_feature_extractor: true |
|
save_path: partly_frozen_splitted_wavlm/save/wav2vec2_hubert_checkpoint |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ctc_lin: &id004 !new:speechbrain.nnet.linear.Linear |
|
|
|
input_size: 1024 |
|
n_neurons: 41 |
|
|
|
log_softmax: !new:speechbrain.nnet.activations.Softmax |
|
apply_log: true |
|
|
|
ctc_cost: !name:speechbrain.nnet.losses.ctc_loss |
|
blank_index: 0 |
|
|
|
modules: |
|
wav2vec2: *id002 |
|
enc: *id003 |
|
ctc_lin: *id004 |
|
model: &id005 !new:torch.nn.ModuleList |
|
- [*id003, *id004] |
|
model_opt_class: !name:torch.optim.Adadelta |
|
lr: 1 |
|
rho: 0.95 |
|
eps: 1.e-8 |
|
|
|
wav2vec_opt_class: !name:torch.optim.Adam |
|
lr: 0.0001 |
|
|
|
lr_annealing_model: &id006 !new:speechbrain.nnet.schedulers.NewBobScheduler |
|
initial_value: 1 |
|
improvement_threshold: 0.0025 |
|
annealing_factor: 0.8 |
|
patient: 0 |
|
|
|
lr_annealing_wav2vec: &id007 !new:speechbrain.nnet.schedulers.NewBobScheduler |
|
initial_value: 0.0001 |
|
improvement_threshold: 0.0025 |
|
annealing_factor: 0.9 |
|
patient: 0 |
|
|
|
|
|
checkpointer: !new:speechbrain.utils.checkpoints.Checkpointer |
|
checkpoints_dir: partly_frozen_splitted_wavlm/save |
|
recoverables: |
|
wav2vec2: *id002 |
|
model: *id005 |
|
scheduler_model: *id006 |
|
scheduler_wav2vec: *id007 |
|
counter: *id008 |
|
train_logger: !new:speechbrain.utils.train_logger.FileTrainLogger |
|
save_file: partly_frozen_splitted_wavlm/train_log.txt |
|
|
|
error_rate_computer: !name:speechbrain.utils.metric_stats.ErrorRateStats |
|
|
|
cer_computer: !name:speechbrain.utils.metric_stats.ErrorRateStats |
|
split_tokens: true |
|
|