Cosine-Distance Virtual Adversarial Training for Semi-Supervised Speaker-Discriminative Acoustic Embeddings.
Florian L. KreyssigPhilip C. WoodlandPublished in: INTERSPEECH (2020)
Keyphrases
- semi supervised
- cosine distance
- supervised learning
- manifold learning
- vector space
- semi supervised learning
- acoustic models
- labeled data
- pairwise
- acoustic features
- euclidean distance
- discriminative training
- unsupervised learning
- prosodic features
- unlabeled data
- speech recognition
- speaker verification
- dimensionality reduction
- hidden markov models
- generative model
- training set
- feature extraction
- edit distance
- active learning
- retrieval model
- neural network
- low dimensional