Search papers, labs, and topics across Lattice.
3
0
4
6
A striking task-dependent robustness gap reveals that while ASR thrives on direct audio retrieval, AQA falters due to bottlenecks in mediated information access.
Genre biases in music evaluation models can lead to significant misalignments with human preferences, but a new training approach can correct this skew.
Aesthetics-guided training enables LeVo 2 to generate songs that not only sound good but also maintain coherence and detail, outperforming existing models.
