xls-r-uyghur-cv8 / README.md
lucio's picture
Update README.md
7fee3c1
|
raw
history blame
4.14 kB
metadata
language:
  - ug
license: apache-2.0
tags:
  - automatic-speech-recognition
  - mozilla-foundation/common_voice_8_0
  - generated_from_trainer
  - ug
  - robust-speech-event
datasets:
  - - mozilla-foundation/common_voice_8_0
model-index:
  - name: XLS-R-300M Uyghur CV8
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Common Voice 8
          type: mozilla-foundation/common_voice_8_0
          args: ug
        metrics:
          - name: Test WER
            type: wer
            value: 30.5
          - name: Test CER
            type: cer
            value: 5.8

XLS-R-300M Uyghur CV8

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

  • Loss: 0.2026
  • Wer: 0.3248

Model description

For a description of the model architecture, see facebook/wav2vec2-xls-r-300m

The model vocabulary consists of the alphabetic characters of the Perso-Arabic script for the Uyghur language, with punctuation removed.

Intended uses & limitations

This model is expected to be of some utility for low-fidelity use cases such as:

  • Draft video captions
  • Indexing of recorded broadcasts

The model is not reliable enough to use as a substitute for live captions for accessibility purposes, and it should not be used in a manner that would infringe the privacy of any of the contributors to the Common Voice dataset nor any other speakers.

Training and evaluation data

The combination of train and dev of common voice official splits were used as training data. The official test split was used as validation data as well as for final evaluation.

Training procedure

The featurization layers of the XLS-R model are frozen while tuning a final CTC/LM layer on the Uyghur CV8 example sentences. A ramped learning rate is used with an initial warmup phase of 2000 steps, a max of 0.0001, and cooling back towards 0 for the remainder of the 9400 steps (100 epochs).

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 0.0001
  • train_batch_size: 16
  • eval_batch_size: 8
  • seed: 42
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 64
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 2000
  • num_epochs: 100.0
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
3.3036 5.32 500 3.2628 1.0
2.9734 10.63 1000 2.5677 0.9980
1.3466 15.95 1500 0.4455 0.6306
1.2424 21.28 2000 0.3603 0.5301
1.1655 26.59 2500 0.3165 0.4740
1.1026 31.91 3000 0.2930 0.4400
1.0655 37.23 3500 0.2675 0.4159
1.0239 42.55 4000 0.2580 0.3913
0.9938 47.87 4500 0.2373 0.3698
0.9655 53.19 5000 0.2379 0.3675
0.9374 58.51 5500 0.2486 0.3795
0.9065 63.83 6000 0.2243 0.3405
0.888 69.15 6500 0.2157 0.3277
0.8646 74.47 7000 0.2103 0.3288
0.8602 79.78 7500 0.2088 0.3238
0.8442 85.11 8000 0.2045 0.3266
0.8335 90.42 8500 0.2038 0.3241
0.8288 95.74 9000 0.2024 0.3280

Framework versions

  • Transformers 4.17.0.dev0
  • Pytorch 1.10.2+cu102
  • Datasets 1.18.3
  • Tokenizers 0.11.0