本文为美国伊利诺伊大学香槟分校（作者：VINAY MADDALI）的学士论文，共27页。

本研究的目的是将个别讯号与混音讯号分开。对同一信号进行多次记录，采用概率算法考虑这些记录共享的共同成分。将乐器演奏的每个音符视为一个单独的变量，并确定在每个录音中观察到这些变量的概率，然后使用聚类算法根据源的数量对这些变量进行分组。取观测变量的概率和它们的归属，并与频谱图进行比较，以分别获得每个声源的声音。在室内环境下对一些信号进行了实验，并给出了提取信号的方法。

The purpose of this research is to separate individual signals from a mixture of audio signals. Multiple recordings of the same signal were taken.A probabilistic algorithm was used that considers the common components that these recordings share. It considers each note played by an instrument as an individual variable and determines the probability of observing those variables in each of the recordings. These variables are then grouped according to the number of sources using a clustering algorithm. The probabilties of observing the variables and where they belong to are taken and compared with the spectrogram to obtain the sound of each source separately. Experiments were conducted on some signals in a room environment and the method has been shown to extract the signals.

1.       引言

2. 已有的工作

3. 推导

4. 聚类

5. 实验结果

6. 局限性与进一步研究

更多精彩文章请关注公众号：