[an error occurred while processing this directive] [an error occurred while processing this directive]
Roger Serwy

Non-Negative Matrix Factorization for Speech Feature Extraction

Roger Serwy, 3/20/2014, 4:30-5:30pm, BI 2369

Non-negative matrix factorization (NMF) provides a way to represent spectrogram-like data using purely additive basis vectors. Several variations of NMF have been proposed in the last decade, including sparseness constraints, using convolution, and incorporating complex phase. NMF on spectrogram data of for nonsense CV tokens may be used for tracking glottal events as well as speech segmentation.

[an error occurred while processing this directive]