Hugging Face's logo

samitizerxu
/
wav2vec2-xls-r-300m-fr

Automatic Speech Recognition
Transformers
PyTorch
French
wav2vec2
common_voice
Generated from Trainer
hf-asr-leaderboard
robust-speech-event
Eval Results
Inference Endpoints
Model card Files Community
1
wav2vec2-xls-r-300m-fr / README.md
anton-l's picture
anton-l HF staff
Upload README.md
bbcb338
preview code
|
raw
history blame contribute delete
No virus
2.76 kB
metadata
language:
  - fr
license: apache-2.0
tags:
  - automatic-speech-recognition
  - common_voice
  - fr
  - generated_from_trainer
  - hf-asr-leaderboard
  - robust-speech-event
datasets:
  - common_voice
model-index:
  - name: wav2vec2-cls-r-300m-fr
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Robust Speech Event - Dev Data
          type: speech-recognition-community-v2/dev_data
          args: fr
        metrics:
          - name: Test WER
            type: wer
            value: 56.62
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Robust Speech Event - Test Data
          type: speech-recognition-community-v2/eval_data
          args: fr
        metrics:
          - name: Test WER
            type: wer
            value: 58.22

wav2vec2-cls-r-300m-fr

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

  • Loss: 0.6521
  • Wer: 0.4330

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: 0.0003
  • train_batch_size: 16
  • eval_batch_size: 8
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • num_epochs: 10.0
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
2.6773 0.8 500 1.3907 0.9864
0.9526 1.6 1000 0.7760 0.6448
0.6418 2.4 1500 0.7605 0.6194
0.5028 3.2 2000 0.6516 0.5322
0.4133 4.0 2500 0.6303 0.5097
0.3285 4.8 3000 0.6422 0.5062
0.2764 5.6 3500 0.5936 0.4748
0.2361 6.4 4000 0.6486 0.4683
0.2049 7.2 4500 0.6321 0.4532
0.176 8.0 5000 0.6230 0.4482
0.1393 8.8 5500 0.6595 0.4403
0.1141 9.6 6000 0.6552 0.4348

Framework versions

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