AI RESEARCH
Sparse Autoencoders for Interpretable Emotion Control in Text-to-Speech
arXiv CS.CL
•
ArXi:2606.01479v1 Announce Type: new Integrating large language models (LLMs) into text-to-speech (TTS) systems has improved speech expressiveness, yet interpretable emotional control remains challenging. Existing approaches primarily rely on external conditioning or global activation steering, offering limited insight into the internal representations underlying emotional control. In this work, we analyze emotion-related variation in the semantic hidden states of LLM-based TTS models using sparse autoencoders (SAEs) to identify sparse latent features.