Whisper Large V3 Turbo Cantonese Yue English

🚀 Transformers

A library for state-of-the-art Natural Language Processing and Speech Processing.

🚀 Quick Start

This project is based on the transformers library, using the openai/whisper-large-v3-turbo model for speech recognition. It has been trained on multiple datasets and achieved a CER of 13.7%.

✨ Features

• Multiple Datasets: Trained on AlienKevin/mixed_cantonese_and_english_speech, mozilla-foundation/common_voice_17_0, and mozilla-foundation/common_voice_11_0.
• Low CER: Achieved a Character Error Rate (CER) of 13.7%.
• Specific Training Args: Defined specific training arguments for better performance.

📦 Installation

If you encounter tokenizer issues during inference, please update your transformers version to >= 4.46.3.

pip install --upgrade transformers==4.46.3

📚 Documentation

Model Information

Training Arguments

per_device_train_batch_size=16,
gradient_accumulation_steps=1,
learning_rate=1e-5,
gradient_checkpointing=True,
per_device_eval_batch_size=16,
generation_max_length=225,

Hardware

• NVIDIA Tesla V100 16GB * 4

FAQ

⚠️ Important Note

If having tokenizer issue during inference, please update your transformers version to >= 4.46.3

📄 License

This project is licensed under the MIT license.