cs.yaml

# Generated 2023-08-03 from:
# /home/salah/new_tunisian_model/hparams/train_tunisian_withwavlm.yaml
# yamllint disable
# ################################
# Model: wav2vec2 + DNN + CTC
# Augmentation: SpecAugment
# Authors: Titouan Parcollet 2021
# ################################

seed: 1994
__set_seed: !!python/object/apply:torch.manual_seed [1234]
output_folder: results/non_semi_final_stac
wer_file: !ref <output_folder>/wer.txt
save_folder: !ref <output_folder>/save
train_log: !ref <output_folder>/train_log.txt



# Data files
data_folder: junk  # e.g, /localscratch/cv-corpus-5.1-2020-06-22/fr
train_tsv_file: junk/train.tsv                # Standard CommonVoice .tsv files
dev_tsv_file: junk/dev.tsv                # Standard CommonVoice .tsv files
test_tsv_file: junk/test.tsv                # Standard CommonVoice .tsv files
accented_letters: true

csv_folder: /gpfsscratch/rech/nou/uzn19yk/switched_data/extended_clean/
train_csv: !ref <csv_folder>/train.csv
valid_csv: !ref <csv_folder>/dev.csv
test_csv:
        - all_tests/cs_test.csv
        - all_tests/stac_test.csv

# We remove utterance slonger than 10s in the train/dev/test sets as
# longer sentences certainly correspond to "open microphones".
avoid_if_longer_than: 13.0
avoid_if_shorter_than: 0.5

# Training parameters
number_of_epochs: 20
lr: 0.0002
lr_weights: 0.01
sorting: ascending
auto_mix_prec: False
sample_rate: 16000
language_modelling: True
ngram_lm_path: arpas/pluslanguages_everything.arpa 

# With data_parallel batch_size is split into N jobs
# With DDP batch_size is multiplied by N jobs
# Must be 3 per GPU to fit 32GB of VRAM
batch_size: 3
test_batch_size: 4

# Dataloader options
dataloader_options:
    batch_size: !ref <batch_size>
    num_workers: 6

test_dataloader_options:
    batch_size: !ref <test_batch_size>
    num_workers: 6

# Model parameters
activation: !name:torch.nn.Sigmoid
dnn_layers: 1
dnn_neurons: 768
freeze_encoder: True

# Outputs
output_neurons: 76  # BPE size, index(blank/eos/bos) = 0

# Functions and classes
#
epoch_counter: !new:speechbrain.utils.epoch_loop.EpochCounter
   limit: !ref <number_of_epochs>

encoder_dim: 3217 
enc: !new:speechbrain.nnet.RNN.LSTM
   input_shape: [Null, Null, !ref <encoder_dim>]
   num_layers: 2
   bidirectional: True
   dropout: 0.2
   hidden_size: 1024

ctc_lin: !new:speechbrain.nnet.linear.Linear
   input_size: 2048
   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:
   enc: !ref <enc>
   ctc_lin: !ref <ctc_lin>

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

model_opt_class: !name:torch.optim.Adam
   lr: !ref <lr>

weights_opt_class: !name:torch.optim.Adam
   lr: !ref <lr_weights>

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

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

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

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

blank_index: 0
unk_index: 1


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