Conference Paper

Noise-robust automatic speech recognition using mainlobe-resilient time-frequency quantile-based noise estimation

Dept. of Electron. Eng., Hong Kong Chinese Univ., Shatin, China
DOI: 10.1109/ISCAS.2004.1328774 Conference: Circuits and Systems, 2004. ISCAS '04. Proceedings of the 2004 International Symposium on, Volume: 3
Source: IEEE Xplore


In standard speech recognition systems in which training data are clean speech, the presence of background noise in received signal can severely deteriorate the recognition performance. This paper presents a simple noise-robust speech recognition system based on a modified noise spectral estimation method called mainlobe-resilient time-frequency quantile-based noise estimation (M-R T-F QBNE), which focuses on the mainlobes at harmonic frequencies. We estimate the global signal-to-noise ratio (SNR) and select a recognition model, which is best matched to the SNR operating range. Experimental results show that the recognition accuracy of the proposed recognition system is higher than that of the AURORA2 clean training baseline by 23%. Compared to multicondition training, the proposed method achieves comparable recognition accuracy.

Download full-text


Available from: Siu Wa Lee, May 22, 2015
  • [Show abstract] [Hide abstract]
    ABSTRACT: This paper presents a novel adaptive wavelet denoising system for speech enhancement. The procedures of the standard wavelet thresholding are improved by optimizing the adaptive algorithms. The combination of the quantile based noise estimate and the posterior SNR based threshold adjuster serves to balance the effects of noise removal and speech preservation. In order to improve the final perceptual quality, a musical noise suppression algorithm and a TEO based silent segment smoothing module are also incorporated into the system. Simulation experiments are designed to demonstrate the capability of the proposed system.
    No preview · Conference Paper · May 2008