search

clovaai / voxceleb_trainer

In defence of metric learning for speaker recognition
MIT License
1.02k stars 272 forks source link

SpeakerNet: score = -1 * numpy.mean(dist) #119

Closed zengxinch closed 3 years ago

zengxinch commented 3 years ago

why should we multiply -1 ?

ukemamaster commented 3 years ago

To convert distance into similarity, i suppose.