多基音检测算法研究-通信与信息系统专业论文.docx

万方数据 万方数据 ABSTRACT Pitch frequency of speech signals is a very important parameter, widely used in speech and audio signal processing fields, such as auditory scene analysis, single channel blind source separation, temperament analysis. At present, for single pitch detection of single speech, we have gotten a lot of algorithms whose performances are excellent, but the pitch detection becomes difficult when several speeches appear at the same time. In this paper, the single pitch and multiple pitch detection algorithms are firstly analyzed and studied. Based on the analysis of characteristics of speech singles and mixed speech signals, a new multiple pitch detection algorithm based on harmonic peaks extraction is proposed. In the algorithm, speeches are processed in frequency domain. Harmonic components of the signal can be gotten by selecting the spectral peaks, and then a statistical method is used to get the octave relationship from the harmonic components. Finally, pitch frequencies of two signals before mixture are obtained. In addition, a voiced/unvoiced decision technique is used to reduce the computational complexity of the algorithm. The new algorithm has been implemented with C language. Test results have shown that, for single speech, the detection accuracy of voiced speech is 81.8% and that of unvoiced speech is 90.2%; for mixed speech, the detection accuracy of voiced speech is 80.7% and that of unvoiced speech is 90.8%. Keyword: Multiple pitch Detection Auditory Model Sinusoidal Model Voiced/Unvoiced decision Harmonic Component 目录 第一章 绪论 ....................................................................................................1 1.1 研究背景和意义.................................................................................................. 1 1.2 基音检测技术发展概况..................................................................................... 1 1.3 本文的主要内容和章节安排............................................................................. 4 第二章 基音检测