A feature extraction method using subband based periodicity and aperiodicity decomposition with noise robust frontend processing for automatic speech recognition.
Kentaro IshizukaTomohiro NakataniPublished in: Speech Commun. (2006)
Keyphrases
- subband
- automatic speech recognition
- noisy environments
- feature vectors
- wavelet packet
- filter bank
- low frequency
- speech recognition
- multiscale decomposition
- high frequency
- wavelet transform
- multiresolution
- image compression
- wavelet coefficients
- speech signal
- subband coding
- wavelet decomposition
- bit rate
- low pass
- high pass
- wavelet domain
- hidden markov models
- subband decomposition
- discrete wavelet transform
- broadcast news
- image coding
- energy compaction
- quadrature mirror
- multiscale
- feature set
- directional information
- conversational speech
- frequency domain
- wavelet coding
- similarity measure
- perfect reconstruction
- noise reduction
- image features
- noise model
- image coder
- geometric distortions
- computationally efficient
- machine learning
- linear prediction
- additive noise
- compression ratio
- image denoising
- denoising