Fish Audio Launches S2.1 Pro, Supporting 83 Languages

Fish Audio has launched S2.1 Pro, a production-grade voice model optimized for real-time conversational speech, moving away from the traditional script-reading approach. This model is available through the Fish Audio API, and a free tier for development and testing is also supported.

The core strengths of S2.1 Pro are its low latency and broad language support. Its key features are as follows:

• Ultra-low latency performance: It records a time-to-first-audio latency of approximately 90ms on standard calls, enabling natural conversation.
• Multilingual support: It supports 83 languages while maintaining a single voice identity.
• Fine-grained emotion control: By inputting bracket tags directly within the text, instructions such as whispering or nervous laughter can be executed without separate configuration.
• High-performance cloning: With just a short sample of 10-30 seconds, it captures the speaker's tone and style without additional fine-tuning.

This model can be easily integrated into agent workflows through MCP and Agent-skill support, making it useful for developers building voice agents, phone systems, or long-form audio pipelines. It was also built on the technical achievements of the open-source Fish Speech lineup, which has recorded over 20,000 stars on GitHub.