A trainable speech enhancement technique based on mixture models for speech and noise.
Ilyas PotamitisNikos FakotakisGeorge KokkinakisPublished in: INTERSPEECH (2003)
Keyphrases
- speech enhancement
- mixture model
- noisy environments
- noise reduction
- signal to noise ratio
- speech signal
- noisy speech
- single channel
- speech recognition
- gaussian mixture model
- em algorithm
- linear prediction
- vocal tract
- probabilistic model
- generative model
- background noise
- language model
- model selection
- wiener filter
- additive noise
- expectation maximization
- multi channel
- unsupervised learning
- maximum likelihood
- smoothing algorithm
- automatic speech recognition
- probability density function
- bayesian information criterion
- edge detection
- non stationary