Using VTLN matrices for rapid and computationally-efficient speaker adaptation with robustness to first-pass transcription errors.
Shakti Prasad RathSrinivasan UmeshAchintya Kumar SarkarPublished in: INTERSPEECH (2009)
Keyphrases
- speaker adaptation
- computationally efficient
- automatic speech recognition
- speech recognition
- maximum likelihood
- speech recognition systems
- speech recognizer
- speaker independent
- hidden markov models
- speech signal
- computer vision
- speaker identification
- prediction error
- image acquisition
- language model
- pattern recognition
- broadcast news