Neural "spike rate spectrum" as a noise robust, speaker invariant feature for automatic speech recognition.
T. V. SreenivasG. V. KiranA. G. KrishnaPublished in: INTERSPEECH (2004)
Keyphrases
- automatic speech recognition
- noisy environments
- spectral subtraction
- speech recognition
- speech signal
- invariant features
- speaker verification
- hidden markov models
- broadcast news
- spoken words
- neural network
- speaker identification
- word error rate
- conversational speech
- speech retrieval
- recognition errors
- acoustic features
- speech synthesis
- speech sounds
- power spectrum
- noise reduction
- gray level
- speaker adaptation
- speech corpus
- language model
- neuronal networks
- input image