Wav2vec2 Xls R 300m Zh HK Lm V2

🚀 Wav2Vec2 XLS-R 300M Cantonese (zh-HK) LM

This is an automatic speech recognition model based on the XLS-R architecture. It fine - tunes Wav2Vec2 - XLS - R - 300M on the zh-HK subset of the Common Voice dataset. A 5 - gram Language model, trained on multiple PyCantonese corpora, is added to enhance its performance. The model is part of the Robust Speech Challenge Event organized by HuggingFace.

🚀 Quick Start

The model is trained using HuggingFace's PyTorch framework. All necessary training scripts can be found in the Files and versions tab, and the Training metrics are logged via Tensorboard. For N - gram language model training, follow the blog post tutorial provided by HuggingFace.

✨ Features

• Based on XLS - R Architecture: Leveraging the power of the XLS - R architecture for better performance in automatic speech recognition.
• Fine - tuned on Common Voice: Fine - tuned on the zh-HK subset of the Common Voice dataset to adapt to Cantonese.
• Language Model Integration: A 5 - gram language model trained on PyCantonese corpora is added to improve recognition accuracy.

📦 Installation

No specific installation steps are provided in the original document.

💻 Usage Examples

No code examples are provided in the original document.

📚 Documentation

Model

Evaluation Results

Without Language Model

With Language Model

Training procedure

Training hyperparameters

• learning_rate: 0.0001
• 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: 100.0
• mixed_precision_training: Native AMP

Training results

Disclaimer

⚠️ Important Note

Consider the biases from pre - training datasets that may affect the model's results.

Authors

The model was trained and evaluated by Wilson Wongso. All computation and development are done on OVH Cloud.

Framework versions

• Transformers 4.17.0.dev0
• Pytorch 1.10.2+cu102
• Datasets 1.18.4.dev0
• Tokenizers 0.11.0

🔧 Technical Details

No specific technical details are provided in the original document.

📄 License

The model is released under the Apache 2.0 license.