Real-time DSP-based audio pattern recognition system

Abstract

Advances in digital signal processing has brought about new topics of concern specially in the real-time analysis and recognition of certain audio signal patterns. Correlation and wavelet analysis techniques used in template pattern recognition requires a great deal of computation time and therefore can not be implemented in real-time. Variations in amplitude or audio volume also impose a problem if direct correlation is used. In order to avoid this problem, the audio pattern recognition algorithm is made amplitude-independent by operating in the frequency domain. Assuming that the plot between the dominant frequency and time (or the pitch plot) of a particular audio signal is unique, a signature pitch plot can then be acquired. Templates of these frequency plots could then be used to identify these audio patterns as their signature.