marinone94's picture
update datasets version
fb5ea5a
|
raw
history blame
4.38 kB
metadata
language:
  - sv-SE
license: cc0-1.0
tags:
  - automatic-speech-recognition
  - mozilla-foundation/common_voice_7_0
  - sv
  - generated_from_trainer
  - robust-speech-event
  - model_for_talk
datasets:
  - mozilla-foundation/common_voice_7_0
model-index:
  - name: XLS-R-300M-voxrex - Swedish
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Common Voice 7
          type: mozilla-foundation/common_voice_7_0
          args: sv-SE
        metrics:
          - name: Test WER
            type: wer
            value: 18.89
          - name: Test CER
            type: cer
            value: 6.63
      - 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: 30.65
          - name: Test CER
            type: cer
            value: 13.56

This model is a fine-tuned version of KBLab/wav2vec2-large-voxrex on the MOZILLA-FOUNDATION/COMMON_VOICE_7_0 - SV-SE dataset. It achieves the following results on the evaluation set:

  • Loss: 0.2201
  • Wer: 0.1778

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
3.1522 1.45 500 3.1290 1.0
2.9576 2.91 1000 2.9633 1.0
1.9853 4.36 1500 0.8902 0.6104
1.5867 5.81 2000 0.4793 0.3664
1.4608 7.27 2500 0.3816 0.3095
1.3496 8.72 3000 0.3415 0.2783
1.3058 10.17 3500 0.3072 0.2519
1.2533 11.63 4000 0.2877 0.2381
1.2535 13.08 4500 0.2791 0.2320
1.2273 14.53 5000 0.2726 0.2282
1.2083 15.99 5500 0.2638 0.2212
1.1606 17.44 6000 0.2531 0.2174
1.1545 18.89 6500 0.2468 0.2109
1.1344 20.35 7000 0.2494 0.2050
1.1173 21.8 7500 0.2447 0.1980
1.1081 23.26 8000 0.2428 0.1998
1.1023 24.71 8500 0.2329 0.1951
1.0923 26.16 9000 0.2388 0.1962
1.0798 27.61 9500 0.2363 0.1944
1.0769 29.07 10000 0.2342 0.1913
1.0672 30.52 10500 0.2250 0.1875
1.0735 31.97 11000 0.2305 0.1874
1.0628 33.43 11500 0.2291 0.1851
1.0451 34.88 12000 0.2263 0.1856
1.0299 36.34 12500 0.2257 0.1834
1.0368 37.79 13000 0.2230 0.1808
1.0322 39.24 13500 0.2231 0.1833
1.0451 40.7 14000 0.2197 0.1817
1.0304 42.15 14500 0.2241 0.1813
1.0102 43.6 15000 0.2233 0.1795
1.0135 45.06 15500 0.2200 0.1794
1.014 46.51 16000 0.2207 0.1779
1.0071 47.96 16500 0.2205 0.1784
0.9729 49.42 17000 0.2204 0.1777

Framework versions

  • Transformers 4.16.0.dev0
  • Pytorch 1.10.1+cu102
  • Datasets 1.17.1.dev0
  • Tokenizers 0.11.0