anton-l's picture
anton-l HF staff
Upload README.md
0c5ac12
metadata
language:
  - sl
license: apache-2.0
tags:
  - automatic-speech-recognition
  - mozilla-foundation/common_voice_7_0
  - generated_from_trainer
  - sl
  - robust-speech-event
  - model_for_talk
  - hf-asr-leaderboard
datasets:
  - mozilla-foundation/common_voice_7_0
model-index:
  - name: XLS-R-300M - Slovenian
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Common Voice 7
          type: mozilla-foundation/common_voice_7_0
          args: sl
        metrics:
          - name: Test WER
            type: wer
            value: 18.97
          - name: Test CER
            type: cer
            value: 4.534
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Robust Speech Event - Dev Data
          type: speech-recognition-community-v2/dev_data
          args: sl
        metrics:
          - name: Test WER
            type: wer
            value: 55.048
          - name: Test CER
            type: cer
            value: 22.739
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Robust Speech Event - Test Data
          type: speech-recognition-community-v2/eval_data
          args: sl
        metrics:
          - name: Test WER
            type: wer
            value: 54.81

wav2vec2-large-xls-r-300m-slovenian

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA-FOUNDATION/COMMON_VOICE_7_0 - SL dataset. It achieves the following results on the evaluation set:

  • Loss: 0.2093
  • Wer: 0.1907

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 7e-05
  • 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: 1000
  • num_epochs: 100.0
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
1.785 12.5 1000 0.7465 0.6812
0.8989 25.0 2000 0.2495 0.2732
0.7118 37.5 3000 0.2126 0.2284
0.6367 50.0 4000 0.2049 0.2049
0.5763 62.5 5000 0.2116 0.2055
0.5196 75.0 6000 0.2111 0.1910
0.4949 87.5 7000 0.2131 0.1931
0.4797 100.0 8000 0.2093 0.1907

Framework versions

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