Specmurt Analysis of Multi-Pitch Music Signals with Adaptive Estimation of Common Harmonic Structure

This paper describes a multi-pitch analysis method using specmurt analysis with iterative estimation of the quasioptimal common harmonic structure function. Specmurt analysis (Sagayama et al., 2004) is based upon the idea that superimposed harmonic structure pattern can be expressed as a convolution of two components, a fundamental frequency distribution and a ‘common harmonic structure’ function if each underlying tone component has similar harmonic structure pattern. As proved in our previous work (Sagayama et al., 2004) inappropriate common structure function leads to inaccurate analysis results. The iterative algorithm proposed in this paper automatically chooses a proper structure, which results in finding concurrent multiple fundamental frequencies and reduces the dependency on heuristically chosen initial common harmonic structure. The experimental evaluation showed promising results.