Noise and Metadata Sensitive Bottleneck Features for Improving Speaker Recognition with Non-Native Speech Input.
Yao QianJidong TaoDavid Suendermann-OeftKeelan EvaniniAlexei V. IvanovVikram RamanarayananPublished in: INTERSPEECH (2016)
Keyphrases
- speaker recognition
- metadata
- mel frequency cepstral coefficients
- speaker identification
- gaussian mixture model
- speaker verification
- vector quantization
- probabilistic neural network
- feature extraction
- speech recognition
- feature set
- input data
- speech signal
- neural network
- feature space
- signal to noise ratio
- noise reduction
- low level
- extracted features
- pattern classification
- maximum likelihood
- feature vectors