@svpino: This is the first audio model I've seen that lets you edit individual words. Text-to-Speech models have gotten really g…
Summary
Fish Audio 推出了 Drama 3 语音模型，支持音频修复（audio inpainting），可以只重新生成音频中的单个词或短语，其余部分保持不变，并支持 83+ 种语言。
View Cached Full Text
Cached at: 10/12/26, 03:30 PM
This is the first audio model I’ve seen that lets you edit individual words.
Text-to-Speech models have gotten really good lately, but they still make mistakes, and I hate having to redo an entire voiceover because one word is wrong.
@FishAudio just released Drama 3, and it supports audio inpainting.
Basically, you can regenerate a word or phrase inside your audio while keeping everything else unchanged.
The workflow is simple:
- Generate your voiceover
- Choose the word or phrase you want to replace
- Generate the replacement and keep the rest of the take
It works surprisingly well!
By the way, you can do this in 83+ languages!
Thanks to the Fish Audio team for partnering with me on this post.
Similar Articles
fishaudio/drama-3
Fish Audio 发布新一代文本转语音模型 Drama 3，采用 35B 参数 MoE 架构（每 token 激活 3B），支持最多 16 个说话人的多角色长文本生成、语音克隆与自然语言控制，并可在 SGLang-Omni 上流式部署。
Fish Audio Drama 3
Fish Audio 推出其最新 AI 语音模型 Drama 3，支持通过自然语言控制情感、节奏、耳语、笑声和停顿，可生成多角色场景、有声书与游戏对话，覆盖 83 种语言，并以社区许可证开放权重供自托管。
Drama 3 (Website)
Fish Audio 发布了 Drama 3 预览版文本转语音模型，支持用自然语言、括号内标签、音效与局部标记来精确指导每一句台词的表演方式，并可通过 API 调用。
Fish Audio Drama 3 Technical Report
Fish Audio 发布开源语音生成模型 Fish Audio Drama 3 的技术报告：基于 35B 参数混合专家骨干（每 token 激活 3B），支持多参考语音克隆、自然语言导演指令、多轮修订、多人场景配音、编辑式填充、长音频与流式合成，单张 H200 推理速度达 271 帧/秒。
@rohanpaul_ai: Really cool, open weight, TTS model, just dropped. Fixing one bad word in a TTS clip usually means regenerating the who…
Fish Audio 发布了开源权重的 TTS 模型 Drama 3，号称全球最具情感表现力和可控性的语音合成模型，支持在句子中对单个单词进行风格和语气微调。