Unsupervised adaptation of a denoising autoencoder by Bayesian Feature Enhancement for reverberant asr under mismatch conditions.
Jahn HeymannReinhold Haeb-UmbachPavel GolikRalf SchlüterPublished in: ICASSP (2015)
Keyphrases
- denoising
- image processing
- speech signal
- image denoising
- natural images
- data driven
- automatic speech recognition
- sufficient conditions
- unsupervised learning
- total variation
- noisy images
- speech recognition
- supervised learning
- bayesian estimation
- image features
- denoising methods
- image enhancement
- posterior probability
- computer vision
- bayesian inference
- maximum likelihood
- markov random field
- feature set
- semi supervised
- unsupervised manner
- restricted boltzmann machine
- adaptation process
- pointwise mutual information