Unsupervised Training for Deep Speech Source Separation with Kullback-Leibler Divergence Based Probabilistic Loss Function.
Masahito TogamiYoshiki MasuyamaTatsuya KomatsuYu NakagomePublished in: CoRR (2019)
Keyphrases
- loss function
- kullback leibler divergence
- information theoretic
- pairwise
- mutual information
- support vector
- supervised learning
- distance measure
- probability density function
- generative model
- audio features
- marginal distributions
- training samples
- unsupervised learning
- semi supervised
- probabilistic model
- speech recognition
- nearest neighbor
- bayesian framework
- hidden markov models