StepAudio 2.5 TTS

StepAudio 2.5 TTS is a text-to-speech model developed by StepFun that integrates contextual understanding directly into the speech generation pipeline. Designed to move beyond static, tag-matching speech synthesizers, the model combines Global Context and Inline Context dual-level controls with zero-shot voice cloning capabilities, allowing generated audio to closely mimic human vocal delivery and performance nuances.

The model introduces two primary entry points for controlling expressiveness and tone. The global instruction parameter allows users to set the overall emotional register and character persona for an entire segment, while inline controls using parentheses within the text input apply fine-grained, per-sentence directions. This dual structure enables precise regulation of rhythm, emotion, and style without requiring manual data collection or complex rule engines.

Supported modalities include text input mapped to various audio formats like MP3, WAV, FLAC, and Opus. It is built to serve scenarios ranging from expressive audiobooks and emotional companionship applications to voice assistants and video dubbing, providing stable performance across multiple languages.