Multiple fundamental frequency estimation assisted by audio watermarking

Abstract

The success of computer segregation of sound sources from a single‐channel mixture often relies on the estimation of multiple fundamental frequencies. Instead of solving the problem directly, this presented work utilizes audio watermarks to assist sound source segregation. Here, individual sources are assumed to be available for watermark embedding before mixing. Each source’s short‐time spectral peaks are aligned to frequency gridpoints labeled with a digital ID [Liu and Smith, ‘‘Audio Watermarking Based on Sinusoidal Analysis and Synthesis,’’ International Symposium on Musical Acoustics, Nara, Japan (2004)]. A modified sound source is sinusoidally synthesized afterwards. In this sense, each source is embedded with a watermark. In an experiment, two sound sources are watermarked with two sets of gridpoints P and Q, respectively. Points in P and Q interleave and are spaced 10 cents of a semitone apart. The watermarked sources are then linearly added. To demix them, sinusoidal trajectories found in the mix...