Conference Paper

Speech enhancement for non-stationary noise environment by adaptivewavelet packet

Dept. of Electr. & Comput. Eng., Hanyang Univ.
DOI: 10.1109/ICASSP.2002.5743779 Conference: Acoustics, Speech, and Signal Processing, 2002. Proceedings. (ICASSP '02). IEEE International Conference on, Volume: 1
Source: IEEE Xplore


We consider the non-stationary or colored noise estimation by wavelet thresholding method. First, we propose node dependent thresholding for adaptation in colored or non-stationary noise. Next, we suggest a noise estimation method based on spectral entropy using histogram of intensity instead of estimation method based on median absolute deviation (MAD). We use a modified hard thresholding to alleviate time-frequency discontinuities. The proposed methods are evaluated on various noise conditions - white Gaussian noise, car interior noise, F-16 cockpit noise, pink noise, speech babble noise. We compare our proposed methods with the conventional one with level dependent thresholding based on MAD

1 Follower
20 Reads
  • Source
    • "Also some other thresholding functions were proposed in Chang et al. (2002) and Sheikhzadeh and Abutalebi (2001). "
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper, we propose a new speech enhancement system using the wavelet thresholding algorithm. The basic wavelet thresholding algorithm has some defects including the assumption of white Gaussian noise (WGN), malfunction in unvoiced segments, bad auditory quality, etc. In the proposed system, we introduce a new algorithm which does not require any voiced/unvoiced detection system. Also, in this proposed method adaptive wavelet thresholding and modified thresholding functions are introduced to improve the speech enhancement performance as well as the automatic speech recognition (ASR) accuracy. A new voice activity detector (VAD) was designed to update noise statistics in the proposed speech enhancement system when facing to the colored and non-stationary noises. The proposed method was evaluated on several speakers and under various noise conditions including white Gaussian noise, pink noise, and multi-talker babble noise. The SNR and ASR results show that the new method highly improves the performance of speech enhancement algorithm based on the wavelet thresholding.
    Full-text · Article · Aug 2006 · Speech Communication
  • Source
    • "In such methods, choice of a good threshold and thresholding function for reduction of noise in the wavelet domain is the challenging subject[7] [10] [11] [12]. Recently, a novel approach for noise reduction using the wavelet thresholding has been proposed by Donoho[7]. "
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper we propose a new approach for speech enhancement. The method used to remove the noise components is a combination of two methods: Wavelet de-noising and spectral subtraction. The idea is to apply the spectral subtraction to wavelet approximations and details coefficients. A new parameter for spectral subtraction in unvoiced speech frames is introduced and the existing power factor in spectral subtraction method is improved. Also, for reduction of musical noise, we propose to use iterative Wiener filtering. Experimental results demonstrate that the proposed speech enhancement algorithm is very promising.
    Full-text · Article · Jan 2004
  • Source
    • "Usually, the wavelet-based methods are developed for removing white Gaussian noise corruption [1]. Chang et al [2] proposed the node dependent thresholding for the adaptation of WCs in colored or non-stationary noise. They suggested a noise estimation method based on spectral entropy using histogram of intensity instead of median absolute deviation. "
    [Show abstract] [Hide abstract]
    ABSTRACT: There are few works on the problem of heavy noise corruption in wavelet-based speech enhancement. In this paper, a new method is introduced to adapt the weighting function for wavelet coefficients (WCs) in each subband. The idea is based on that the variance of WCs in speech-dominated frames is larger than the variance of WCs in noise-dominated frames. We can define a weighting function for WCs in each subband so that WCs are preserved in speech-dominated frames and reduced in noise-dominated frames. Then a weighting function in terms of WC's variance is derived. The experimental results show that the proposed method is more robust than that of SNR adjusted speech enhancement system.
    Preview · Conference Paper · Jan 2003
Show more