Search papers, labs, and topics across Lattice.
2
0
3
3
Achieve state-of-the-art universal speech enhancement with UniPASE, a model that minimizes linguistic hallucinations by converting degraded waveforms into clean phonetic representations.
Generative and predictive models can be fused to achieve state-of-the-art speech enhancement, outperforming single-branch approaches in robustness and perceptual quality.