An Improved Array Steering Vector Estimation Method and Its Application in Speech Enhancement
- PDF / 777,532 Bytes
- 8 Pages / 600 x 792 pts Page_size
- 4 Downloads / 167 Views
An Improved Array Steering Vector Estimation Method and Its Application in Speech Enhancement Zhu Liang Yu Center for Signal Processing, Nanyang Technological University, Singapore 639798 Email: [email protected]
Meng Hwa Er School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 639798 Email: [email protected] Received 30 April 2004; Revised 22 February 2005 We propose a robust microphone array for speech enhancement and noise suppression. To overcome target signal cancellation problem of conventional beamformes caused by array imperfections or reverberation effects, the proposed method adopts arbitrary transfer function relating each microphone and target speech signal as array channel model. This is achieved in two ways. First, we propose a method to estimate the array steering vector (ASV) by means of exploiting the nonstationarity of speech signal to combat stationary noise and interference. Next, with the estimated ASV, a robust matched-filter-(MF-) array-based generalized sidelobe canceller (MF-GSC) is constructed to enhance the speech signal and suppress noise/interference. In addition, it also has the capability to reduce the reverberation effects of the acoustic enclosure. Numerical results show that the proposed method demonstrates high performance even in adverse environments. Keywords and phrases: adaptive microphone array, matched-filter array, generalized sidelobe canceller, noise suppression, speech enhancement, nonstationarity.
1.
INTRODUCTION
Speech enhancement and noise suppression have received increasing interest in speech-related applications in adverse environments. Conventional single-channel speech enhancement methods, such as spectral subtraction [1], do not provide sufficient improvement to the speech signal, especially when noise is strong. In the past few decades, microphone array has been proposed as a promising technique for speech enhancement. It uses the signals captured by multimicrophones, which are distributed at different positions, to exploit the spatial-temporal information of the target signal, interferences, and noise for the purpose of improving the signal-to-noise ratio (SNR) by suppressing background noise and interferences. Beamforming is the key technique in microphone array for speech enhancement and noise suppression. Many beamforming methods [2, 3, 4, 5, 6] have been proposed in literature. Among them, the most famous algorithm for wideband beamforming is the constrained minimum power adaptive beamformer proposed by Frost [2], also called Frost beamformer. It is capable of satisfying certain desired frequency response in the looking direction while minimizing the output
noise power through constrained minimization of the total output power. Griffiths and Jim [3] reconstructed the Frost beamformer into the generalized sidelobe canceller (GSC). It transforms the constrained optimization problem in Frost beamformer into an unconstrained one and, consequently, improves the convergence performance. To improve the robustness of Frost beamform
Data Loading...
