Source identification and manipulation in stereo music recordings using frequency‐domain signal processing

A short‐time frequency domain framework for source identification, separation, and manipulation in stereo music recordings is presented. Using a simplified model of the stereo mix, a similarity measure between the short‐time fourier transforms (STFTs) of the input signals is computed to identify time‐frequency regions occupied by each source based on the panning coefficients assigned to it during the mix. Individual sources are identified and manipulated by clustering time‐frequency components with a given panning coefficient and frequency range. After modification, an inverse STFT is used to synthesize a time‐domain processed signal. Applications of the technique to source suppression, enhancement and repanning will be described, and audio demonstrations will be presented to illustrate the results.