First model version

checkpoint_best.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:13ee473f13953e0d9053f39367251669e2a006af3d99f5b007dfe2a79b35f851
+size 1140805865

dict.ltr.txt

@@ -0,0 +1,89 @@
+| 1425817
+e 1178319
+n 633984
+a 534584
+i 432845
+t 426822
+d 368464
+r 363974
+o 362103
+s 243285
+l 240045
+h 194637
+g 193570
+k 170494
+m 163029
+u 157460
+v 141792
+j 130640
+w 111117
+z 97895
+b 97538
+p 82921
+c 75002
+f 51103
+' 20301
+X 7489
+é 7286
+y 5108
+è 5063
+- 4778
+ë 3361
+x 2022
+q 682
+ï 493
+ü 264
+ö 157
+à 139
+ê 80
+ç 39
+1 38
+á 38
+ä 35
+2 34
+32 30
+áx 16
+38 15
+16 13
+35 12
+4 11
+ô 10
+â 10
+21 10
+& 10
+ñ 8
+314 6
+î 5
+12 5
+24 5
+19 5
+17 5
+40 4
+10 4
+130 4
+52 4
+5 3
+8 3
+20 3
+3 3
+Ö 3
+í 3
+6 2
+04 2
+313 2
+ó 2
+201 1
+67 1
+499 1
+7 1
+45 1
+198 1
+25 1
+902 1
+xq 1
+3xx 1
+66 1
+ù 1
+ò 1
+Å 1
+ú 1

finetuning_config.yaml

@@ -0,0 +1,65 @@
+# @package _group_
+
+common:
+  memory_efficient_fp16: true
+  log_format: json
+  log_interval: 100
+  model_parallel_size: 1
+
+checkpoint:
+  no_epoch_checkpoints: true
+  best_checkpoint_metric: wer
+  save_dir: /esat/spchtemp/scratch/jponcele/selfsupervised_exps/result/finetune_VW_base_all
+
+task:
+  _name: audio_pretraining
+  data: /users/spraak/jponcele/BenchmarkingSS/data/cgn_phone_10ms_w2v2_all
+  normalize: true  #false
+  labels: ltr
+  segments: true
+  max_length: 800000
+
+dataset:
+  num_workers: 6
+  batch_size: 4
+  max_tokens: 32000000
+  skip_invalid_size_inputs_valid_test: true
+  valid_subset: test
+  data_buffer_size: 2
+
+distributed_training:
+  ddp_backend: legacy_ddp
+  distributed_world_size: 1
+
+criterion:
+  _name: ctc
+  zero_infinity: true
+
+optimization:
+  max_update: 500000
+  lr: [0.00003]
+  sentence_avg: true
+  update_freq: [4]
+
+optimizer:
+  _name: adam
+  adam_betas: (0.9,0.98)
+  adam_eps: 1e-08
+
+lr_scheduler:
+  _name: tri_stage
+  phase_ratio: [0.1, 0.4, 0.5]
+  final_lr_scale: 0.05
+
+model:
+  _name: wav2vec_ctc
+  w2v_path: /esat/spchtemp/scratch/jponcele/selfsupervised_exps/result/pretrain_w2v2_cgn-unsup-VW_base/checkpoint_74_250000.pt
+  apply_mask: true
+  mask_prob: 0.65
+  mask_channel_prob: 0.5
+  mask_channel_length: 64
+  layerdrop: 0.1
+  activation_dropout: 0.1
+  feature_grad_mult: 0.0
+  freeze_finetune_updates: 0
+

pretraining_config.yaml

@@ -0,0 +1,63 @@
+# @package _group_
+
+common:
+  memory_efficient_fp16: true
+  log_format: json
+  log_interval: 100
+
+checkpoint:
+  save_interval_updates: 10000
+  keep_interval_updates: 1
+  no_epoch_checkpoints: true
+  save_dir: /esat/spchtemp/scratch/jponcele/selfsupervised_exps/result/pretrain_w2v2_cgn-unsup-VW_base
+
+task:
+  _name: audio_pretraining
+  data: /users/spraak/jponcele/BenchmarkingSS/data/cgn_unsup_VW_w2v2
+  max_sample_size: 250000
+  min_sample_size: 4000
+  segments: true
+  normalize: true
+
+dataset:
+  num_workers: 6
+  #batch_size: 4
+  max_tokens: 1400000
+  skip_invalid_size_inputs_valid_test: true
+  valid_subset: test
+  data_buffer_size: 1  #2
+  required_batch_size_multiple: 1  #default=8
+
+distributed_training:
+  distributed_world_size: 1
+  ddp_backend: legacy_ddp
+
+criterion:
+  _name: wav2vec
+  infonce: true
+  log_keys: ["prob_perplexity","code_perplexity","temp"]
+  loss_weights: [0.1, 10]
+
+optimization:
+  max_update: 400000
+  lr: [0.0005]
+  update_freq: [32]
+
+optimizer:
+  _name: adam
+  adam_betas: (0.9,0.98)
+  adam_eps: 1e-06
+  weight_decay: 0.01
+
+lr_scheduler:
+  _name: polynomial_decay
+  warmup_updates: 50000
+
+model:
+  _name: wav2vec2
+  quantize_targets: true
+  final_dim: 256
+  encoder_layerdrop: 0.05
+  dropout_input: 0.1
+  dropout_features: 0.1
+  feature_grad_mult: 0.1