| Property | Value |
|---|---|
| Model Type | Speech Recognition |
| Base Architecture | Whisper Large-v3 Turbo |
| License | MIT |
| Author | erax-ai |
| Training Data | 600,000 samples (1000 hours) |
What is EraX-WoW-Turbo-V1.1-CT2?
EraX-WoW-Turbo-V1.1-CT2 is a high-performance speech recognition model that builds upon Whisper Large-v3 Turbo, optimized specifically for Vietnamese and 10 other languages. The model achieves remarkable speed through CTranslate2 integration, processing 30 seconds of audio in approximately 350ms, making it suitable for real-time applications.
Implementation Details
The model leverages CTranslate2 library for enhanced performance, offering up to 2.5x speedup compared to standard implementations. It supports GPU acceleration with FP16 precision and includes built-in VAD (Voice Activity Detection) filtering for improved accuracy.
- Multilingual support for 11 languages including Vietnamese (all 8 regions), English, Chinese, Cantonese, Indonesian, Korean, Japanese, Russian, German, French, and Dutch
- Word Error Rate (WER) of approximately 12% across major languages
- Optimized for real-world audio conditions including noise handling
- Seamless integration with popular Python libraries including pydub, silero-vad, and faster-whisper
Core Capabilities
- Real-time transcription with minimal latency
- Robust handling of regional accents and dialects
- Efficient noise handling and audio preprocessing
- Support for multiple audio input formats
- Integration with voice assistance systems
- Media subtitling and accessibility tools
Frequently Asked Questions
Q: What makes this model unique?
The model's distinctive feature is its optimization for Vietnamese language processing while maintaining high performance across multiple languages, combined with unprecedented speed through CTranslate2 integration.
Q: What are the recommended use cases?
The model excels in real-time transcription scenarios, live captioning, media subtitling, voice assistants, accessibility tools, and language learning applications. It's particularly effective for applications requiring low-latency response times and high accuracy across multiple languages.