Unsupervised Beamforming Based on Multichannel Nonnegative Matrix Factorization for Noisy Speech Recognition.
Kazuki ShimadaYoshiaki BandoMasato MimuraKatsutoshi ItoyamaKazuyoshi YoshiiTatsuya KawaharaPublished in: ICASSP (2018)
Keyphrases
- speech recognition
- nonnegative matrix factorization
- noisy environments
- speech enhancement
- speech signal
- negative matrix factorization
- matrix factorization
- hidden markov models
- speech recognizer
- data representation
- language model
- least squares
- semi supervised
- linear prediction
- automatic speech recognition
- speech synthesis
- spectral clustering
- pattern recognition
- unsupervised learning
- principal component analysis
- speaker identification
- multi channel
- supervised learning
- speech recognition systems
- missing data
- objective function
- feature selection
- noise reduction
- original data
- document clustering
- n gram
- em algorithm
- information retrieval systems
- active learning
- multiscale
- clustering algorithm
- computer vision