Poly-InstructTTS Enables Expressive Speech Synthesis via Natural Language
August 24, 2026
Poly-InstructTTS utilizes a prompt-free GPT with attribute-based thinking tokens and a flow-matching module to map open-ended instructions to speech. The model was trained on a 1,000-hour corpus covering over 1,000 fine-grained emotions and styles for improved instruction adherence.
HOW THIS AFFECTS YOU
●
builderYou can implement fine-grained emotional control in TTS applications using natural language prompts.
●
designerThis enables more nuanced and expressive voice personas for user interfaces.