fishaudio/drama-3 is an open-weight, directable text-to-speech model released on Hugging Face for multi-speaker, emotionally expressive audio generation and editing. Designed for dramatic or narrative use, it provides fine-grained control over delivery and speaker identity, enabling producers to generate acted lines, emotional inflections, and multi-voice scenes. The model targets audio drama workflows - synthesis, voice replacement, and scene-level direction - rather than simple neutral TTS, and ships in safetensors format suitable for local inference and integration into editing pipelines.
The repository metadata shows fishaudio/drama-3 as a public Hugging Face model (pipeline_tag: text-to-speech) with downloads and community engagement (839 downloads, 207 likes) and recent updates in October 2026. Tags and labels highlight multi-speaker, audio-editing, emotional-speech-generation and support for a very large set of languages (English, Chinese, Japanese, Korean, Spanish, Portuguese, Arabic, Russian, French, German and many more). Additional technical hints include an arXiv identifier, a nonstandard license, references to bsgemm-moe and region:us, and availability as safetensors, indicating ready-to-run weights for experimentation and production use.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.