摘要语音分离问题源自声学领域经典的“鸡尾酒会效应”，指的是人在复杂的人声环境中，能够选定并追踪某一感兴趣的声音，而让计算机拥有人的这种能力，便是语音分离要解决的问题。如今，语音分离技术在语音识别、人机对话、人声检索以及声源定位等领域拥有宽广的运用前景。本文阐述了语音分离相关的基础理论、BSS的数学建模以及三种常用的ICA算法。主要工作如下：

本设计使用Kinect for Windows采集的真实语音进行人工线性瞬时混合后的信号作为待分离信号进行语音分离技术的仿真实验，并对Infomax算法、JADE算法、FastICA算法的性能进行比较和分析，实验的结果表明这三种算法在不考虑噪声的线性瞬时混合信号情况下有着良好的分离性能。80704

毕业论文关键词  语音分离  盲源信号分离  独立分量分析  线性瞬时混合

毕业设计说明书外文摘要

Title  Speech Separation based on KINECT

Abstract Speech separation problem stems from the classic "cocktail party effect" in the field of acoustics, in other words, people can select and track a specific sound in the complex environment of the human voice。 The major task of speech separation is to make the computer own this ability。 Nowadays, speech separation technology has broad application prospects in terms of speech recognition, man-machine dialog and sound source positioning,etc。 This thesis describes the basic concepts of speech separation, mathematical model of blind source signal separation and three commonly used algorithms of independent component analysis(ICA)。 The main work can be summarized as follows: 

This design uses real voice signals collected by Kincet and artificially mix them as linear instantaneous mixture signals for the simulation experiment of speech separation technology。 This design also compares and analyzes the performance of Infomax algorithm, JADE algorithm and FastICA algorithm, the results of experiment indicate that these three algorithms show good performances in the case of linear instantaneous mixed signals separation without considering environmental noise。

Keywords  Speech separation  Blind source signal separation  Indenpent component analysis  Linear instantaneous mixture

目   次

1  绪论 1

1。1 语音分离问题出现的原因 1

1。2 语音分离技术的发展进程 2

1。3 语音分离技术的应用与展望 3

1。4 本文内容安排 4

2  语音分离相关基础知识 5

2。1 语音分离相关基本概念 5

2。2 盲源分离的数学模型 8

2。3盲源分离的特性 9 

3  语音分离的几种算法研究 10 

3。1 语音信号的预处理 10

3。2 语音信号的分离 10

3。3 语音分离算法的性能指标 14

4  语音分离实验与结果分析 16 

4。1 语音分离的实验环境 16

4。2 语音分离的实验步骤 16

4。3 语音分离的实验结果与分析 22

结论 26