Multichannel audio can provide immersive environment the listener by rendering surround audio. But such audio resolution takes more effort on recording tracks from multiple microphones for content production. Multichannel audio synthesize significantly reduces the excessive size requirement of the set of signals for transmission medium.
In multichannel audio synthesis, the instrument enhancement technique plays an important role since it can improve the performance of virtual spot microphones that often used to capture individual instruments. One of the well-developed methods is spectral conversion based Guassian mixture model. By considering the spectral vectors sequence of a given signal as a realization of random vector x defined with probability density function, the problem will become finding a proper transfer function ,which , when applied to input vector x produces a vector close in some sense to target vector, in this case, the selected and enhanced instrument track.
Based on my experiences, the spectral conversion algorithms that have been successfully employed to speech synthesis applications are basically used to achieve the instruments recognition and enhancement. Also, some adaptive algorithms will be proposed to strengthen the robustness and improve the performance of the enhancement.
My main focus in this project is, to implement instruments that recognize processing of several musical pieces that consists of several instruments. Then the validation will be via both signal analysis and listening tests. The expected results of the classification of musical pieces can be observed by system output and human auditory system.
The main subjects are:
Frequency Scale Conversion
Fourier, DCT and Hartley Transforms
Random Number Generation
Probability Density Functions
Vector Distance
Calculate psychometric function
Speech Synthesis and Speech Enhancement
Speech Coding
Signal Processing
Calculates entropy and conditional entropy for discrete and continuous distributions
Voice box Parameters and System Interface
Utility Functions
Functions that return the sin and cos of the angle
Calculates log(sum(exp(x))) without overflow problems
