Regularized nonnegative matrix factorization using Gaussian mixture priors for supervised single channel source separation.
Emad M. GraisHakan ErdoganPublished in: Comput. Speech Lang. (2013)
Keyphrases
- single channel
- source separation
- multi channel
- prior information
- matrix factorization
- independent component analysis
- em algorithm
- negative matrix factorization
- frequency domain
- wiener filter
- expectation maximization
- semi supervised
- least squares
- unsupervised learning
- mixture model
- wavelet decomposition
- supervised learning
- prior knowledge
- bayesian framework
- objective function
- maximum a posteriori
- spectral clustering
- probability density function
- noise reduction
- signal processing
- brain activity
- closed form
- non stationary
- collaborative filtering
- probabilistic model
- image processing