Fish Speech

A self-hosted text-to-speech system clones voices and controls emotion; code and weights use the FISH AUDIO RESEARCH LICENSE.

Fish Speech is a self-hosted text-to-speech system for creators producing narration and developers building voice applications. Its S2 Pro model supports directed speech delivery and dialogue with multiple speakers.

Voice cloning: Reproduce a speaker's voice and speaking style from a short reference recording without additional fine-tuning.
Emotion and delivery: Direct whispers, emphasis, laughter and pauses at specific points in a script, with free-form descriptions for more detailed control.
Multilingual speech: Generate speech in English, Japanese, Chinese, Korean, Spanish, French, German and Arabic, among other languages.
Multi-speaker dialogue: Generate distinct voices from a reference recording containing multiple speakers, and use earlier conversation turns as context for later speech.

Self-hosted streaming uses SGLang. Fish Audio reports GPU performance measurements using an NVIDIA H200. The code and model weights use the custom FISH AUDIO RESEARCH LICENSE.

Fish Audio separately provides a hosted browser service and APIs for speech generation and voice cloning. Transcription, voice agents and a library of user-uploaded voices are part of that cloud offering.