Fish Speech V1.4
| Property | Value |
|---|---|
| License | CC-BY-NC-SA-4.0 |
| Research Paper | arXiv:2411.01156 |
| Languages Supported | 8 (English, Chinese, German, Japanese, French, Spanish, Korean, Arabic) |
| Training Data Size | 700,000 hours |
What is fish-speech-1.4?
Fish Speech V1.4 is a state-of-the-art multilingual text-to-speech (TTS) model that represents a significant advancement in speech synthesis technology. Trained on an impressive 700,000 hours of audio data across eight different languages, it leverages large language models for enhanced multilingual speech synthesis capabilities.
Implementation Details
The model has been trained with a particular focus on English and Chinese, with approximately 300,000 hours of training data for each of these languages. The remaining six languages (German, Japanese, French, Spanish, Korean, and Arabic) each benefit from around 20,000 hours of training data, ensuring robust performance across all supported languages.
- Primary language support: English and Chinese (300k hours each)
- Secondary language support: 20k hours each for German, Japanese, French, Spanish, Korean, and Arabic
- Implementation available on GitHub with demo access through Fish Audio platform
Core Capabilities
- High-quality speech synthesis in 8 different languages
- Advanced multilingual text processing
- Balanced performance across various language pairs
- Research-focused architecture leveraging LLM technologies
Frequently Asked Questions
Q: What makes this model unique?
The model's extensive training data (700k hours) and balanced approach to major languages sets it apart, especially with its deep focus on English and Chinese content. The integration with large language models for text-to-speech synthesis represents a novel approach in multilingual TTS systems.
Q: What are the recommended use cases?
The model is ideal for research purposes and non-commercial applications requiring high-quality multilingual speech synthesis. It's particularly well-suited for applications requiring English or Chinese speech synthesis, given the extensive training in these languages.
