kingabzpro's picture
Update README.md
8034834
|
raw
history blame
3.74 kB
metadata
language:
  - sv-SE
license: apache-2.0
tags:
  - automatic-speech-recognition
  - robust-speech-event
datasets:
  - mozilla-foundation/common_voice_8_0
metrics:
  - wer
  - cer
model-index:
  - name: wav2vec2-large-xls-r-1b-Swedish
    results:
      - task:
          type: automatic-speech-recognition
          name: Speech Recognition
        dataset:
          type: mozilla-foundation/common_voice_8_0
          name: Common Voice sv-SE
          args: sv-SE
        metrics:
          - type: wer
            value: 18.03
            name: Test WER Without LM
            args:
              - learning_rate: 0.000075
              - train_batch_size: 32
              - eval_batch_size: 8
              - seed: 42
              - gradient_accumulation_steps: 4
              - total_train_batch_size: 128
              - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
              - lr_scheduler_type: linear
              - lr_scheduler_warmup_steps: 1000
              - num_epochs: 50
              - mixed_precision_training: Native AMP
          - type: cer
            value: 5.69
            name: Test CER  Without LM
            args:
              - learning_rate: 0.000075
              - train_batch_size: 32
              - eval_batch_size: 8
              - seed: 42
              - gradient_accumulation_steps: 4
              - total_train_batch_size: 128
              - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
              - lr_scheduler_type: linear
              - lr_scheduler_warmup_steps: 1000
              - num_epochs: 50
              - mixed_precision_training: Native AMP

wav2vec2-large-xls-r-1b-Swedish

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on the common_voice dataset. It achieves the following results on the evaluation set:

Without LM

  • Loss: 0.3370
  • Wer: 0.1803
  • Cer: 0.0569

With LM

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 7.5e-05
  • train_batch_size: 32
  • eval_batch_size: 8
  • seed: 42
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 128
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 1000
  • num_epochs: 50
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer Cer
3.1423 5.49 500 0.5523 0.4414 0.1313
0.8615 10.98 1000 0.3877 0.2946 0.0942
0.4848 16.48 1500 0.3580 0.2539 0.0798
0.3538 21.97 2000 0.3391 0.2254 0.0709
0.2879 27.47 2500 0.3392 0.2151 0.0680
0.2466 32.96 3000 0.3687 0.2131 0.0680
0.2146 38.46 3500 0.3551 0.1951 0.0618
0.1916 43.95 4000 0.3601 0.1867 0.0590
0.175 49.45 4500 0.3370 0.1803 0.0569

Framework versions

  • Transformers 4.17.0.dev0
  • Pytorch 1.10.2+cu102
  • Datasets 1.18.2.dev0
  • Tokenizers 0.11.0