xls-r-300m-sv-cv8 / README.md
anton-l's picture
anton-l HF staff
Upload README.md
56c6586
metadata
language:
  - sv-SE
license: apache-2.0
tags:
  - automatic-speech-recognition
  - mozilla-foundation/common_voice_8_0
  - generated_from_trainer
  - sv
  - robust-speech-event
  - hf-asr-leaderboard
datasets:
  - mozilla-foundation/common_voice_8_0
model-index:
  - name: XLS-R-300M - Swedish - CV8 - v2
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Common Voice 8
          type: mozilla-foundation/common_voice_8_0
          args: sv-SE
        metrics:
          - name: Test WER
            type: wer
            value: 17.33
          - name: Test CER
            type: cer
            value: 5.8
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Robust Speech Event - Dev Data
          type: speech-recognition-community-v2/dev_data
          args: sv
        metrics:
          - name: Test WER
            type: wer
            value: 27.01
          - name: Test CER
            type: cer
            value: 12.92

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA-FOUNDATION/COMMON_VOICE_8_0 - SV-SE dataset. It achieves the following results on the evaluation set:

  • Loss: 0.2779
  • Wer: 0.2525

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: 7.5e-05
  • train_batch_size: 8
  • eval_batch_size: 8
  • seed: 42
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 32
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 2000
  • num_epochs: 50.0
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
3.3224 1.37 500 3.3354 1.0
2.9318 2.74 1000 2.9361 1.0000
2.1371 4.11 1500 1.1157 0.8359
1.6883 5.48 2000 0.6003 0.6314
1.5812 6.85 2500 0.4746 0.4725
1.5145 8.22 3000 0.4376 0.4736
1.4763 9.59 3500 0.4006 0.3863
1.4215 10.96 4000 0.3783 0.3629
1.3638 12.33 4500 0.3555 0.3425
1.3561 13.7 5000 0.3340 0.3228
1.3406 15.07 5500 0.3373 0.3295
1.3055 16.44 6000 0.3432 0.3210
1.3048 17.81 6500 0.3282 0.3118
1.2863 19.18 7000 0.3226 0.3018
1.2389 20.55 7500 0.3050 0.2986
1.2361 21.92 8000 0.3048 0.2980
1.2263 23.29 8500 0.3011 0.2977
1.2225 24.66 9000 0.3017 0.2959
1.2044 26.03 9500 0.2977 0.2782
1.2017 27.4 10000 0.2966 0.2781
1.1912 28.77 10500 0.2999 0.2786
1.1658 30.14 11000 0.2991 0.2757
1.148 31.51 11500 0.2915 0.2684
1.1423 32.88 12000 0.2913 0.2643
1.123 34.25 12500 0.2777 0.2630
1.1297 35.62 13000 0.2873 0.2646
1.0987 36.98 13500 0.2829 0.2619
1.0873 38.36 14000 0.2864 0.2608
1.0848 39.73 14500 0.2827 0.2577
1.0628 41.1 15000 0.2896 0.2581
1.0815 42.47 15500 0.2814 0.2561
1.0587 43.83 16000 0.2738 0.2542
1.0709 45.21 16500 0.2785 0.2578
1.0512 46.57 17000 0.2793 0.2539
1.0396 47.94 17500 0.2788 0.2525
1.0481 49.31 18000 0.2777 0.2534

Evaluation Commands

  1. To evaluate on mozilla-foundation/common_voice_7_0 with split test
python eval.py --model_id patrickvonplaten/xls-r-300m-sv-cv8 --dataset mozilla-foundation/common_voice_8_0 --config sv-SE --split test
  1. To evaluate on speech-recognition-community-v2/dev_data
python eval.py --model_id patrickvonplaten/xls-r-300m-sv-cv8 --dataset speech-recognition-community-v2/dev_data --config sv --split validation --chunk_length_s 5.0 --stride_length_s 1.0

Framework versions

  • Transformers 4.17.0.dev0
  • Pytorch 1.10.1+cu113
  • Datasets 1.18.4.dev0
  • Tokenizers 0.10.3