耦合MMSE和WEDM幅度谱估计的语音增强方法
CSTR:
作者:
作者单位:

北方工业大学电子信息工程学院 北京100144

中图分类号:

TP391TN911


Coupled MMSE and WEDM spectral amplitude estimations for speech enhancement
Author:
Affiliation:

College of Electronic and Information Engineering, North China University of Technology, Beijing 100144, China

  • 摘要
  • | |
  • 访问统计
  • |
  • 参考文献 [13]
  • |
  • 相似文献 [20]
  • | | |
  • 文章评论
    摘要:

    为同时在高SNR和低SNR条件下获得较好的语音质量,提出了一种耦合MMSE和WEDM幅度谱估计的语音增强方法。此方法利用Sigmoid映射函数将后验SNR映射到[0,1]之间。然后,根据此映射函数,提出一种自适应的谱恢复增益函数计算方法。在高SNR条件下,为避免语音,所提方法采用较大的映射函数值,从而使MMSE谱估计增益函数为谱恢复增益函数。反之,在低SNR条件下,为有效抑制含噪语音中的噪声,所提方法使用较小的映射函数值,从而选择WEDM谱估计增益函数为谱恢复增益函数。实验结果表明,所提算法性能在客观性能测试方面要优于参考算法。

    Abstract:

    In order to achieve better speech quality both in high SNR (signaltonoise ratio) and low SNR conditions, this paper proposes a coupled MMSE (minimum mean square error) and WEDM (weighted euclidean distortion measure) spectral amplitude estimations method for speech enhancement. This method employs sigmoid mapping function to map a posteriori SNR into [0, 1] range. Thus, according to this mapping function, we develop an adaptive gain function calculation method for spectral restoration. In higher SNR conditions, to prevent speech distortions,the proposed method adopts a larger value of mapping function and makes MMSE spectrum estimation gain function be the gain function for spectral restoration. On the other hand, in lower SNR conditions, to more effectively remove noise from noisy speech, the proposed method uses a smaller value of mapping function and selects the gain function of WEDM spectrum estimation to be the gain function for spectral restoration. Our experiments demonstrate that the proposed method is superior to the reference methods.

    参考文献
    [1]杨金宵, 沈天飞, 滕秋霞. 基于声门激励的语音语速、音量调整方法[J]. 电子测量技术, 2016, 39(2): 7275.
    [2]何侃, 田亚清, 李强, 等. 基LD3320的语音识别智能垃圾桶设计[J]. 国外电子测量技术, 2015, 34(6): 8588.
    [3]PHILIPOSC L. Speech enhancement: theory and practice [M]. CRC, 2007.
    [4]SCALARTP, FILHO J V. Speech enhancement based on a priori signal to noise estimation[C]. In Proceedings of ICASSP, 1996: 629632.
    [5]EPHRAIMY, MALAH D. Speech enhancement using a minimum meansquare error shorttime spectral amplitude estimator[J]. IEEE Transactions on Acoustics, Speech and Signal Processing, 1984, 32(6): 11091121.
    [6]LOTTERT, VARY P. Speech enhancement by maximum a posteriori spectral amplitude estimation using a superGaussian speech model[J]. Eurasip Journal on Applied Signal Processing, 2005(7): 11101126.
    [7]SU Y C, TSAO Y, WU J E, et al. Speech enhancement using generalized maximum posteriori spectral amplitude estimators [C]. In Proceedings of ICASSP, 2013:74677471.
    [8]MCAULAY RJ, QUATIERI T F. Speech analysis/synthesis based on a sinusoidal representation [J]. IEEE Transactions on Acoustics, Speech and Signal Processing, 1986, 34(4): 744–754.
    [9]SRINIVASAN S, SAMUELSSON J, KLEJIN W B. Codebookbased Bayesian speech enhancement for nonstationary environments [J]. IEEE Transactions on Audio, Speech and Language Processing, 2007, 15(2): 441451.
    [10]NAIDUD H R, SRINIVASAN S. Robust Bayesian estimation for contextbased speech enhancement [J]. Eurasip Journal on Audio, Speech and Music Processing, 2014.
    [11]COHENI. Noise estimation by minima controlled recursive averaging for robust speech enhancement [J]. IEEE Signal Process. Letters, 2002, 9(1): 1215.
    [12]QUACKENBUSHS R, BARNWELL T P, CLEMENTS M A. Objective Measures of Speech Quality[M]. Englewood Cliffs, NJ: Prentice Hall, 1988.
    [13]ABRAMSONA, COHEN I. Simultaneous detection and estimation approach for speech enhancement [J]. IEEE Transactions on Audio, Speech and Language Processing, 2007, 15(8): 23482359.
    引证文献
    网友评论
    网友评论
    分享到微博
    发 布
引用本文

韩勇,赵宇红.耦合MMSE和WEDM幅度谱估计的语音增强方法[J].国外电子测量技术,2016,35(10):25-29

复制
分享
文章指标
  • 点击次数:951
  • 下载次数: 1412
  • HTML阅读次数: 0
  • 引用次数: 0
历史
  • 在线发布日期: 2016-11-23
文章二维码
×
《国外电子测量技术》
2025年投稿方式有变