Reproduce “Emotional Prosody Control for Speech Generation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Sarath Sivaprasad, Saiteja Kosgi, Vineet Gandhi
Researchers from IIIT Hyderabad and TCS Research developed an extension to the FastSpeech2 architecture that enables continuous emotional control in synthesized speech using the Arousal-Valence psychological model. The system integrates multi-speaker capabilities while maintaining speech naturalness and effectively expressing a range of emotions.
Key insights
- Integrating the Arousal-Valence space provides a continuous, intuitive, and psychologically grounded approach for emotional control in speech synthesis.
- A two-stage training strategy effectively addresses the challenge of limited emotionally annotated data by leveraging large general datasets for the backbone and smaller, annotated datasets for emotional control.
- Decoupling emotional conditioning from the core encoder-decoder pathway ensures that emotional prosody is modified without compromising the fundamental naturalness and quality of the synthesized speech.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.