Frequency component grouping based sound source extraction from mixed audio signals using spectral analysis

The extraction of the polyphonic sounds from mix of different sources is natural, even to normal human listeners. For this reason, human auditory recognition system is very robust compared to the algorithms used to analyze audio in different computing systems. This work is a frequency domain based approach for sound source extraction from any mixed audio signal to improve the efficiency of audio recognition algorithms. In this work, we propose a new approach to estimate, analyze and extract different melodies from different physical sources without modeling the signal or sources. This process uses one or two microphone(s) based systems to analyze sound signal to be compatible to human hearing. Our work is specifically aiming the extraction of sound sources by matching frequency patterns and calculation of delay times, providing the spatial and temporal data of the signal sources and their outputs.