# Fish Speech

> Inference code and weights for Fish Audio Drama 3 (35B-A3B MoE) and S2-Pro (4B): multi-speaker TTS, voice cloning from 10-30 s of audio, inline emotion tags. Not open source: code and Drama 3 use the Fish Audio Community License (commercial use only under USD 500k annual revenue, no hosted TTS services); S2-Pro weights are research and non-commercial only. Heavy hardware: Drama 3 needs at least 48 GB VRAM on NVIDIA; S2-Pro 24 GB or CPU. Licence updated 13 Oct 2026.

- Repository: https://github.com/fishaudio/fish-speech
- Official docs: https://speech.fish.audio/install/
- Checked: 2026-10-15

## What it replaces
- ElevenLabs: Speech generation, Voice cloning (full). Covers multilingual TTS and voice cloning locally, but commercial use is capped by revenue and hosted TTS/cloning services need a separate licence.
- Descript: Voice cloning, Speech generation (partial). Voice cloning and text-to-speech only; no transcript-based audio or video editing.

## Facts, with sources

| Question | Status | Answer | Source | Checked |
|---|---|---|---|---|
| Runs on my machine? | Confirmed | Yes, CLI, API server and WebUI, or Docker; Drama 3 needs an NVIDIA GPU, S2-Pro also runs on CPU | https://speech.fish.audio/install/ | 2026-10-15 |
| Code licence | No | Fish Audio Community License 1.0 (source-available, not OSI). Free for research and non-commercial use; commercial use only below USD 500,000 annual revenue; hosted TTS services excluded | https://github.com/fishaudio/fish-speech/blob/main/LICENSE | 2026-10-15 |
| Model weights licence | No | Drama 3: Fish Audio Community License (same revenue cap). S2-Pro: Fish Audio Research License, non-commercial only. Note the Drama 3 model card metadata says fish-audio-research-license while its text says Community License. | https://huggingface.co/fishaudio/s2-pro | 2026-10-15 |
| Commercial use | Partial | Only with Drama 3 and only while you and your affiliates earn under USD 500k a year; S2-Pro needs a paid licence for any commercial use; offering a hosted TTS/cloning service always needs a separate licence | https://github.com/fishaudio/fish-speech/blob/main/LICENSE | 2026-10-15 |
| Paid services needed? | Partial | None for local research/non-commercial use; a paid commercial licence is needed above the revenue cap or for hosted services | https://github.com/fishaudio/fish-speech/blob/main/LICENSE | 2026-10-15 |
| Hardware | Confirmed | Drama 3 needs at least 48 GB GPU memory (FP8), NVIDIA only; S2-Pro needs at least 24 GB, NVIDIA GPU or CPU | https://speech.fish.audio/install/ | 2026-10-15 |
| Operating systems | Partial | Linux and WSL documented; native Windows and macOS not documented as supported | https://speech.fish.audio/install/ | 2026-10-15 |
| Setup effort | Confirmed | Medium, pip or uv install with a CUDA extra, then download weights; Docker images available | https://speech.fish.audio/install/ | 2026-10-15 |

GitHub: 33193 stars, 2884 forks, last push 2026-10-14, archived: no.
---
Source: RepoQuarry (https://repoquarry.dev/projects/fish-speech). Data licence: CC BY 4.0. Quotes are from project documentation and belong to their authors.