Unverified50% confidenceReleaseExact time
The Fish team released a brand-new text-to-speech model, Drama 3, using an MoE architecture with 35B total parameters and 3B activated parameters.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/15/2026
First Seen
Valid until: 1/13/2027
Sources
Fish Releases Drama 3: A 35B3A MoE Text-to-Speech Model That Tops the TTS Charts
redditr/LocalLLaMA10/15/2026
Related Entities
Related Claims
UnverifiedFish Audio released Drama 3 in October 2026, a 35-billion-parameter speech generation model, with weights published on Hugging Face.74% similarUnverifiedFish Speech是一个开源的文本转语音模型，其S2 Pro版本支持零样本和少样本声音克隆，只需几秒到几十秒的参考音频即可生成高度相似的合成语音61% similarUnverifiedThe technical pipeline for the audiobook generation experiment consists of three steps: generating text, organizing it into a book, and converting it to speech.58% similarUnverifiedThe official team has extended the series into three derivative versions: Index-Echo (end-to-end speech input and speech-to-text), Index-Humana (speech translation with tone), and a Native-Log model that supports length control.57% similarUnverifiedTencent open-sourced a voice generation and editing model that can modify recorded content using natural language while preserving the melody and original timbre; the base model has 1.5 billion parameters, and a Flash distilled version with four-step inference is also provided, with code and weights publicly released.57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1033882API
curl https://kongchang.com/api/v1/knowledge/claims/1033882MCP
get_claim(id=1033882)