Jing Lu

PhASE-Flow achieves state-of-the-art speech enhancement by leveraging phonetic conditioning in the SSL latent space, outperforming traditional spectral domain methods.

HALO reduces computational costs by 50% without sacrificing performance, revolutionizing lightweight speech enhancement models.

Stimulus reconstruction-based auditory attention decoding from EEG signals is easily fooled: unbalanced datasets inflate decoding accuracy, but a new cross-validation method fixes this.

Achieve state-of-the-art universal speech enhancement with UniPASE, a model that minimizes linguistic hallucinations by converting degraded waveforms into clean phonetic representations.