Reproduce “Multi-interaction TTS toward professional recording reproduction” 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
Hiroki Kanagawa, Kenichi Fujita, Aya Watanabe, Yusuke Ijima
Researchers at NTT developed a text-to-speech (TTS) system that enables iterative refinement of synthesized speech style through textual directions, emulating a director-actor feedback loop. This approach, which includes a novel dataset and a dedicated style refiner module, allows for precise stylistic adjustments without compromising the naturalness of the generated speech.
Key insights
- Iterative textual feedback can effectively refine synthesized speech styles, mirroring professional creative workflows.
- A specialized dataset capturing sequential textual directions and corresponding speech refinements is essential for training interactive TTS models.
- A dedicated style refiner module can be integrated into a backbone TTS system to enable multi-step style adjustments without compromising speech naturalness.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.