Dual-Microphone Applications of Multi-Layer Kalman Filter in Speech Enhancement

論文翻譯標題: 多層卡爾曼濾波器於雙麥克風語音降噪之應用
  • 陳 威廷

學生論文: Master's Thesis


Digital speech signal has been widely used in many fields such as telecommunications video conference and artificial intelligence systems However the speech signal inevitably disturbed by the various sound sources of surroundings in the actual communication environments due to the microphone acoustic-electric conversion often lead to recognizable problems for the speech receiving end So how to reduce the environment noise and maintain low distortion speech signal is a major issue in speech signal processing now and attract a lot of attention This study proposes a method called multilayer Kalman filter design based on two different capability microphones (omnidirectional microphone and unidirectional microphone) in hardware to estimate the speech signal and remove the background noises First the unidirectional microphone which is used to collect the ambient background noises is built up in the rear side and then the first-layer Kalman filter is applied on modeling the state-space model of background noises and estimate accurately background noises simultaneously Using the estimating results of background noises to whiten the ambient sound sources (main sound source and background noise) which are collected by the omnidirectional microphone set up directly to face the main sound source is to get the roughly main sound source model Therefore the second-layer Kalman filter can be applied on the roughly main sound source model to extract more pure main sound source By following the similar filtering process mentioned above the proposed multilayer Kalman filter with the increment of layers can achieve the goal of reducing ambient background noises and maintain low distortion of the main source For testing the robustness of this proposed method several signals which the initial SNRs are all lower than 0dB are created and the qualities of estimated results will be verified by the improved SNR cross-correlation PESQ and spectrogram The filter’s layers and orders will be determined to get the optimal quality of speech signal under the consideration of the practical instant processing ability From the revealed research results the proposed method in this paper can significantly enhance the quality of the speech signal
獎項日期2016 八月 30
監督員Yung-Yu Chen (Supervisor)