A generative-discriminative hybrid approach to multi-channel noise reduction for robust automatic speech recognition.
Hendrik MeutznerShoko ArakiMasakiyo FujimotoTomohiro NakataniPublished in: ICASSP (2016)
Keyphrases
- multi channel
- noise reduction
- noisy environments
- automatic speech recognition
- discriminative training
- single channel
- speech recognition
- edge preserving
- speech enhancement
- signal to noise ratio
- wiener filter
- edge detection
- hidden markov models
- speech signal
- noise level
- median filter
- generative model
- noise removal
- word error rate
- broadcast news
- image enhancement
- conversational speech
- neural network
- speech corpus
- spontaneous speech
- language model
- feature selection