A hybrid approach to electrolaryngeal speech enhancement based on spectral subtraction and statistical voice conversion.
Kou TanakaTomoki TodaGraham NeubigSakriani SaktiSatoshi NakamuraPublished in: INTERSPEECH (2013)
Keyphrases
- speech enhancement
- spectral subtraction
- speech recognition
- speech signal
- noisy environments
- signal to noise ratio
- noise reduction
- noisy speech
- single channel
- linear prediction
- vocal tract
- hidden markov models
- hearing aids
- power spectrum
- speech synthesis
- background noise
- wiener filter
- smoothing algorithm
- pattern recognition
- multi channel
- independent component analysis
- prediction error
- spectral analysis
- prior information
- image restoration
- denoising
- neural network