Discriminative PLDA training with application-specific loss functions for speaker verification.
Johan RohdinSangeeta BiswasKoichi ShinodaPublished in: Odyssey (2014)
Keyphrases
- application specific
- loss function
- speaker verification
- general purpose
- pairwise
- support vector
- noisy environments
- learning to rank
- audio visual
- loss minimization
- training set
- language identification
- neural network
- feature extraction
- multilayer perceptron
- convex loss functions
- file system
- pattern recognition
- information retrieval
- noise reduction
- face detection
- nearest neighbor