Fish Audio

Fish Audio provides state-of-the-art TTS and voice cloning; users can create high-fidelity voice copies from just 10–30 seconds of audio. The models support 80+ languages and fine-grained emotional control through natural language tags such as [whisper], [laughing], or [excited tone]. It enables multi-character dialogues with interruptions and realistic turn-taking, making it suitable for podcasts, audiobooks, and conversational AI. Its main strength lies in open weights: the S2 series models and fine-tuning code are publicly available, allowing local deployment. The latest S2.1 Pro model achieved 56.3 characters per second in independent benchmarks and is offered through a free API (model name “s2.1-pro-free”) with no hard usage cap, dramatically lowering the barrier for developers. On the downside, voice cloning introduces ethical risks like impersonation, and the platform currently lacks visible identity-verification or content-moderation measures. Local serving requires GPU resources and technical expertise, and the long-term reliability and concurrency limits of the free API remain unverified at scale.