Search papers, labs, and topics across Lattice.
Nara Institute of Science and Technology
3
0
4
Injecting fine-grained non-verbal cues into TTS can almost perfectly convey sadness (98.3%) and significantly boosts expressiveness overall, despite slight naturalness trade-offs.
A single architecture now unlocks both fast GAN-based and diffusion-based neural vocoders, achieving state-of-the-art speed and quality while drastically reducing training time.
Ditch the training data: this intelligibility-guided approach fuses noisy and enhanced speech for robust ASR without needing a separate neural predictor.
