Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
6
0
7
Achieving precise control over accent intensity in TTS systems could revolutionize speech synthesis for low-resource languages.
OscillaTTS achieves a breakthrough in modeling expressive speech dynamics, outperforming traditional methods in capturing rapid prosodic changes.
Dravidian languages can achieve up to 30% lower WER with novel decoder enhancements that balance attention sources and reinforce token predictions.
Skip expensive human ratings: this hierarchical multimodal model accurately predicts human perception of AI-dubbed content quality using only audio, video, and text inputs.
Achieve startlingly realistic face aging that preserves identity and background consistency, thanks to a novel diffusion-based framework that finely balances aging cues with structural details.
Forget catastrophic forgetting: EW-DETR enables object detectors to learn new classes and adapt to new domains without prior data, boosting overall performance by 57%.