Wav2vec2 Large Xls R 300m Cantonese

🚀 XLS-R-300M - Chinese_HongKong (Cantonese)

This is a fine - tuned automatic speech recognition model based on the XLS - R - 300M architecture. It is trained on the MOZILLA - FOUNDATION/COMMON_VOICE_8_0 - ZH - HK dataset, aiming to provide high - quality speech recognition services for Cantonese in Hong Kong.

🚀 Quick Start

This model is a fine - tuned version of [facebook/wav2vec2 - xls - r - 300m](https://huggingface.co/facebook/wav2vec2 - xls - r - 300m) on the MOZILLA - FOUNDATION/COMMON_VOICE_8_0 - ZH - HK dataset. It achieves the following results on the evaluation set:

• Loss: 1.4848
• Wer: 0.8004

📚 Documentation

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: 32
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• 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: 500
• num_epochs: 100.0
• mixed_precision_training: Native AMP

Training results

Evaluation Commands

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

python eval.py --model_id ivanlau/wav2vec2 - large - xls - r - 300m - cantonese --dataset mozilla - foundation/common_voice_8_0 --config zh - HK --split test --log_outputs

2. To evaluate on speech - recognition - community - v2/dev_data

python eval.py --model_id ivanlau/wav2vec2 - large - xls - r - 300m - cantonese --dataset speech - recognition - community - v2/dev_data --config zh - HK --split validation --chunk_length_s 5.0 --stride_length_s 1.0 --log_outputs

Framework versions

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

📄 License

This project is licensed under the Apache 2.0 license.