TR2012-002

Factorial Models for Noise Robust Speech Recognition


    •  Hershey, J.R.; Rennie, S.J.; Le Roux, J., "Factorial Models for Noise Robust Speech Recognition" in Techniques for Noise Robustness in Automatic Speech Recognition, Virtanen, T. and Singh, R. and Raj, B., Eds., chapter 12, Wiley, November 2012.
      BibTeX Download PDF
      • @incollection{Hershey2012nov,
      • author = {Hershey, J.R. and Rennie, S.J. and {Le Roux}, J.},
      • title = {Factorial Models for Noise Robust Speech Recognition},
      • booktitle = {Techniques for Noise Robustness in Automatic Speech Recognition},
      • year = 2012,
      • editor = {Virtanen, T. and Singh, R. and Raj, B.},
      • chapter = 12,
      • month = nov,
      • publisher = {Wiley},
      • url = {https://www.merl.com/publications/TR2012-002}
      • }
  • MERL Contact:
  • Research Areas:

    Artificial Intelligence, Speech & Audio


TR Image
Figure 12.10 Illustration of the linearization procedure in VTS for a single frequency.

Noise compensation techniques for robust automatic speech recognition (ASR) attempt to improve system performance in the presence of interference from acoustic signals in the environment other than the speech being recognized. In feature-based noise compensation, which includes speech enhancement, the features extracted from the noisy speech signal are modified before being sent to the recognizer by attempting to remove the effects of noise on the speech features. These methods are discussed in Chapter 12. Model compensation approaches, in contrast, are concerned with extending the acoustic model of speech to account for the effects of noise. A taxonomy of different approaches to noise compensation is depicted in Figure 1.1, which serves as a road map to the present discussion.