计算机工程与应用 ›› 2012, Vol. 48 ›› Issue (15): 113-118.

• 数据库、信号与信息处理 • 上一篇    下一篇

一种基于粒子滤波的混合声源跟踪算法

蔡卫平,殷  侠,徐  健   

  1. 九江职业技术学院 电气工程学院,江西 九江 332007
  • 出版日期:2012-05-21 发布日期:2012-05-30

Hybrid algorithm for speech source tracking based on particle filtering

CAI Weiping, YIN Xia, XU Jian   

  1. School of Electrical Engineering, Jiujiang Vocational and Technical College, Jiujiang, Jiangxi 332007, China
  • Online:2012-05-21 Published:2012-05-30

摘要: 为了提高噪声和混响环境中说话人跟踪的精度,提出一种基于粒子滤波的混合声源跟踪算法。根据接收信号信噪比变化较大的特点,该算法使用相位变换加权的可控响应功率定位函数来计算每帧信号的粒子状态观测值,利用其方差将接收信号帧分为高信噪比和低信噪比两种。对于高信噪比帧,仍采用该定位函数构造的似然函数来评价粒子权重,对于低信噪比帧,则采用常规可控波束形成定位函数构造的似然函数来评价粒子权重。仿真结果表明,在平均信噪比较高的条件下,该算法的跟踪性能与传统算法接近;在平均信噪比低于10 dB,混响时间大于200 ms的条件下,跟踪误差比传统算法减少20%~30%。

关键词: 声源跟踪, 粒子滤波, 定位函数

Abstract: A hybrid speech source tracking algorithm based on particle filtering is presented to improve tracking accuracy in noisy and reverberant environment. According to the characteristic of the received signal whose Signal-to-Noise Ratio(SNR) changes in a wide range, the measurement of the particle set for every signal frame is computed using a localization function of the Steered Response Power-phase Transform(SRP-PHAT), and its variance is used to determine the SNR level of the received signal frame in this algorithm. For a high-SNR frame, the particle weight is measured by SRP-PHAT, while, for a low-SNR frame, it is measured by the conventional steered beamformer. Simulation results show that the tracking accuracy of both the proposed algorithm and the traditional algorithm is similar in the condition of high average SNR. When the average SNR is lower than 10 dB, and the reverberation time is more than 200 ms, the tracking error of the proposed algorithm decreases by 20%~30%, compared to the traditional algorithm.

Key words: speech source tracking, particle filter, localization function