README.md

metadata

language:
  - cs
  - hsb
  - pl
  - sk
  - sl
  - multilingual
license: apache-2.0
tags:
  - automatic-speech-recognition
  - generated_from_trainer
  - hf-asr-leaderboard
  - mozilla-foundation/common_voice_8_0
  - robust-speech-event
  - xlsr-fine-tuning-week
datasets:
  - mozilla-foundation/common_voice_8_0
model-index:
  - name: wav2vec2-xls-r-300m-west-slavic-cv8
    results:
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: Common Voice 8
          type: mozilla-foundation/common_voice_8_0
          args: cs
        metrics:
          - type: wer
            value: 53.5
            name: Test WER
          - type: cer
            value: 14.7
            name: Test CER
          - type: wer
            value: 81.7
            name: Test WER
          - type: cer
            value: 21.2
            name: Test CER
          - type: wer
            value: 60.2
            name: Test WER
          - type: cer
            value: 15.6
            name: Test CER
          - type: wer
            value: 69.6
            name: Test WER
          - type: cer
            value: 20.7
            name: Test CER
          - type: wer
            value: 73.2
            name: Test WER
          - type: cer
            value: 23.2
            name: Test CER
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: Robust Speech Event - Dev Data
          type: speech-recognition-community-v2/dev_data
          args: cs
        metrics:
          - type: wer
            value: 84.11
            name: Test WER
          - type: wer
            value: 65.3
            name: Test WER
          - type: wer
            value: 88.37
            name: Test WER
          - type: wer
            value: 87.69
            name: Test WER
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: Robust Speech Event - Test Data
          type: speech-recognition-community-v2/eval_data
          args: cs
        metrics:
          - type: wer
            value: 75.99
            name: Test WER
          - type: wer
            value: 72
            name: Test WER
          - type: wer
            value: 89.08
            name: Test WER
          - type: wer
            value: 87.89
            name: Test WER

wav2vec2-xls-r-300m-west-slavic-cv8

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the Common Voice 8 dataset of five similar languages with similar scripts: Czech, Slovak, Polish, Slovenian and Upper Sorbian. Training and validation sets were concatenated and shuffled.

Evaluation set used for training was concatenated from the respective test sets and shuffled while limiting each language to at most 2000 samples. During training, cca WER 70 was achieved on this set.

Evaluation script

python eval.py --model_id comodoro/wav2vec2-xls-r-300m-west-slavic-cv8 --dataset mozilla-foundation/common_voice_8_0 --split test --config {lang}

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0003
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 50
• mixed_precision_training: Native AMP

Framework versions

• Transformers 4.16.0.dev0
• Pytorch 1.10.1+cu102
• Datasets 1.18.3
• Tokenizers 0.11.0