Search papers, labs, and topics across Lattice.
Affiliation:
2
0
Standard accent classification evaluations overstate model robustness by up to 40% due to speaker overlap; speaker-disjoint, cross-corpus testing reveals a stark performance drop.
