Search papers, labs, and topics across Lattice.
Hangzhou Dianzi University
4
0
5
GSU-DBNet outperforms classical ANN methods in speech enhancement while using only a fraction of the parameters.
Time-unconditional generative speech enhancement achieves better quality and efficiency by sidestepping the constraints of temporal conditioning.
Achieving 98.73% AUC, KFC-KWS redefines user-defined keyword spotting by expertly fusing keyframe and utterance representations to tackle phonetic confusion.
Proprietary MLLMs generate visually appealing code from Figma designs, but fall short on crucial aspects like responsive layouts and maintainability, revealing a surprising gap in their practical utility.
