Decomposition and recognition of a multi-channel audio source using matching pursuit algorithm

Performing real time note detection of multiple audio sources is a difficult task due to the complex nature of audio signals. Typically notes are close in pitch and the harmonic overtones of each note are strongly interlaced, thus preventing standard filtering techniques. In this paper, we develop signal-optimized harmonic atom matching pursuit algorithm, which will be used to decompose an audio signal in terms of elementary waveforms called harmonic atoms. Matching pursuit is an iterative algorithm that correlates the signal against a dictionary set of Gabor atoms and searches for the atoms that exhibit the highest level of correlation. The resulting matched atoms allow for accurate and efficient detection of a single note or multiple notes played simultaneously. In this paper prior knowledge of the signal is used to calculate a parameter set to avoid a costly search over the complete dictionary. The parameter set defined for our proposed algorithm includes time-location, decay rate, frequency, scale and phase, which are calculated at the onset of each note played. The onset (transient) is detected by using a fourth order digital difference then an appropriate interval and optimal window atom is selected. This optimized algorithm is demonstrated through synthesized analysis example. The application of this algorithm in multi-source audio signal can also be demonstrated.