vutankiet2901's picture
update model card README.md
b12fdd1
|
raw
history blame
2.28 kB
metadata
language:
  - ja
license: apache-2.0
tags:
  - automatic-speech-recognition
  - mozilla-foundation/common_voice_8_0
  - generated_from_trainer
model-index:
  - name: wav2vec2-large-xlsr-53-ja
    results: []

wav2vec2-large-xlsr-53-ja

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the MOZILLA-FOUNDATION/COMMON_VOICE_8_0 - JA dataset. It achieves the following results on the evaluation set:

  • Loss: 0.4214
  • Wer: 0.3375
  • Cer: 0.1587

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
  • gradient_accumulation_steps: 2
  • 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 Cer
4.7776 4.73 1500 2.9540 0.9772 0.8489
1.9076 9.46 3000 0.7146 0.5371 0.2484
1.507 14.2 4500 0.5843 0.4689 0.2196
1.3742 18.93 6000 0.5286 0.4321 0.1988
1.2776 23.66 7500 0.5007 0.4056 0.1870
1.2003 28.39 9000 0.4676 0.3848 0.1802
1.1281 33.12 10500 0.4524 0.3694 0.1720
1.0657 37.85 12000 0.4449 0.3590 0.1681
1.0129 42.59 13500 0.4266 0.3423 0.1617
0.9691 47.32 15000 0.4214 0.3375 0.1587

Framework versions

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