Fish Audio S1

Fish Audio S1 is a frontier text-to-speech audio foundation model trained on over two million hours of multilingual audio data. Developed by Fish Audio, the model jointly models semantic and acoustic information to minimize synthesis artifacts and reduce word error rates. S1 is available in multiple variants, including a full-featured 4 billion parameter flagship model and a distilled s1-mini variant designed for lightweight environments.

Key Capabilities and Control

The model supports advanced zero-shot voice cloning and fine-grained expressive control across numerous languages. By leveraging instruction-following annotations derived from specialized audio captioning, users can guide the delivery using emotion and tone markers wrapped in parentheses, such as indicators for specific moods or vocal styles. This design facilitates highly natural speech synthesis suitable for dynamic narration and interactive applications.