Pickdrama-3: Supports 80 Languages, Preserving Emotional Expression in TTS

This is the latest model from Fish Audio, going beyond simple speech synthesis to support emotional expression and audio editing. It naturally speaks over 80 languages, including Korean, English, Japanese, and Chinese, and can implement scenarios where multiple characters converse through its Multi-speaker functionality.

It adopts the bsgemm-moe architecture for efficient inference and runs on the fish_omni library. By applying context-appropriate intonation and emotion to text input, it overcomes the monotony of traditional TTS and is specialized for generating natural dramatic dialogue.

It is suitable for global projects requiring multilingual support, or for audiobook and game development that needs unique voices and emotions for each character. Distributed in safetensors format on Hugging Face, it can be easily integrated into existing pipelines to secure high-quality voice data.