Fish Audio Drama-3
#11 en Synthèse vocale (TTS)Fish Audio · v3 · depuis 2026-10-13 · 2× · vu le 16 oct. 2026
38
Momentum
Fish Audio Drama 3 is a text-to-speech model from Fish Audio, released on October 13, 2026. It uses a 35B-parameter mixture-of-experts architecture with about 3B active parameters per token and supports 83 languages. The weights are on Hugging Face under the Fish Audio Community License v1.0 (not an unrestricted open-source license), and it is also accessible via web app, mobile apps and API.
Historique du momentum
20.07.18.10.
Fonctionnalités
| Real-Time Streaming | Yes: synthesizes long-form and streaming audio; inference via SGLang-Omni; streaming text input new in Drama 3 |
| Latency | Time to first audio 88.9 ms (single stream, 1x H200, per model card) |
| License | Fish Audio Community License v1.0: free for research and non-commercial use; free commercially below US$500,000 annual revenue; hosting as third-party API requires separate license |
| Platform | Web (fish.audio), iOS and Android apps, API (POST /v1/tts with header model: drama-3); weights on Hugging Face, code on GitHub |
| Price | API list price $50 per 1M characters; launch offer $10 per 1M characters (Oct 13–Nov 12, 2026). Deviating: OpenRouter lists $2 per 1M UTF-8 bytes. |
| Release Date | October 13, 2026 |
| Languages | 83 |
| Voice Cloning | Yes: clones voices from multiple references; voices via reference_id in the API |