Advanced Comb Filtering for Robust Speech Recognition
This paper proposes a speech enhancement scheme that leads to significant improvements in recognition performance when used in the Automatic Speech Recognition (ASR) front-end. The proposed approach is based upon adaptive comb filtering. While adaptive comb filtering reduces noise components remarkably, it is rarely effective in reducing non-stationary noises due to its uniformly distributed frequency response. This paper proposes an advanced comb filtering technique that adjusts its spectral magnitude to the original speech, based on the gain modification function, an Minimum Mean Squared Error (MMSE) estimator.
KeywordsAdvanced comb filtering Gain modification function Minimum mean squared error Robust speech recognition
This work was supported by the NAP (National Agenda Project) of the Korea Research Council of Fundamental Science and Technology.
- 1.Frazier RH, Samsam S (1976) Enhancement of speech by adaptive filtering. In: Proceedings of IEEE International Conference on ASSP, Philadelphia. PA, pp 251–253Google Scholar
- 5.Malah D, Cox RV, Accardi AJ (1999) Tracking speech-presence uncertainty to improve speech enhancement in nonstationary noise environments. In: Proceedings of ICASSP, pp 201–204Google Scholar
- 6.Hirsch HG, Pearce D (2000) The AURORA experimental framework for the performance evaluations of speech recognition systems under noisy conditions. Proc ICSLP 4:29–32Google Scholar
- 7.Young S, Evermmana G, Gales M, Hain T et al (2006) The HTK Book for HTK Version 3.4., Cambridge University Engineering Department,CambridgeGoogle Scholar