Study On Key Techniques In Speech Enhancement With Microphone Array

Speech signal processing is one of the kernel techniques applied in many fields such as modern communications, multimedia techniques, and artificial intelligence systems. Generally speaking, the recorded signals are inevitably interfered by the noises from the environment, the reverberation of the rooms and interfering speech sources from other speakers. As a pre-processing method, the speech enhancement technique is an effective way to suppress the interference. Though traditional single microphone techniques can make some effort on noise reduction, due to the information that can be used is solely based on temporal/spectral information about the recorded signals, the performance is unsatisfactory in high reverberation environment or the desired speaker signal is corrupted by other speakers. To solve these problems, spatial and temporal/spectral information can be jointly exploited by using microphone array.Existing array systems have been used in a number of applications including teleconferencing, speech recognition, speaker identification, speech acquisition in an automobile environment, sound capture in reverberant enclosures, sound source localization, and hearing aid devices. These applications can be summarized into two main groups: sound source localization and speech enhancement based on microphone array. As one of the important applications of the microphone array, the speech enhancement with microphone array involves in three key techniques: time delay estimation, voice activity detection and speech enhancement methods. The work in this paper focuses on these three key techniques, which are discussed as follows:The performance of the time delay estimation method based on crosspower-spectrum phase has been analyzed. This method is proposed based on the ideal model of time delay estimation, so it is just suitable for the higher signal-to-noise ratio and lower reverberation case. To make this method applicable in lower signal-to-noise ratio and middle or high reverberation environments, the performance of the method is analyzed and the reason for the poor time delay estimation performace is found. Then two modified weighting functions are given to improve the performance of time delay estimation in lower signal-to-noise ratio and middle or high reverberation condition.In the reverberant environments, the adaptive eigenvalue decomposition based time delay estimation method is studied. Through analyzing its performance, a conclusion can be drawn that the method is only suitable for the mild noise environments. To solve this problem,...