README.md

metadata

language:
  - ja
license: apache-2.0
tags:
  - automatic-speech-recognition
  - mozilla-foundation/common_voice_8_0
  - generated_from_trainer
  - robust-speech-event
  - ja
datasets:
  - common_voice
model-index:
  - name: XLS-R-300-m
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Common Voice 8
          type: mozilla-foundation/common_voice_8_0
          args: ja
        metrics:
          - name: Test WER
            type: wer
            value: 94.91
          - name: Test CER
            type: cer
            value: 23.32

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

• Loss: 0.5351
• Wer: 2.6188

Kanji are converted into Hiragana using the pykakasi library during training and evaluation. The model can output both Hiragana and Katakana characters.

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: 48
• eval_batch_size: 8
• seed: 42
• 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
4.221	4.5	1000	4.1195	2.4024
2.3597	9.01	2000	1.1024	2.7618
1.8795	13.51	3000	0.7498	2.5885
1.7143	18.02	4000	0.6539	2.5976
1.6025	22.52	5000	0.5989	2.6034
1.5403	27.03	6000	0.6035	2.6946
1.4773	31.53	7000	0.5647	2.5558
1.4228	36.04	8000	0.5477	2.5676
1.3801	40.54	9000	0.5413	2.6192
1.3558	45.05	10000	0.5343	2.6575
1.3298	49.55	11000	0.5349	2.6274

Framework versions

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

Evaluation Commands

1. To evaluate on mozilla-foundation/common_voice_8_0 with split test

python ./eval.py --model_id AndrewMcDowell/wav2vec2-xls-r-300m-japanese --dataset mozilla-foundation/common_voice_8_0 --config ja --split test --log_outputs