Sound Clustering Synthesis Using Spectral Data

This paper presents a new sound synthesis method utilizing the features of transitions contained in an existing sound, using spectral data obtained through Short-Time Fourier Transform (STFT) analysis. In this method, spectra obtained from each instantaneous sound are considered as multivariate data, and placed in a vector space, where an evaluation of distances between vectors is performed. As a result, it is possible to detect the occurrences of similarity between analyzed sounds. Clustering and labeling these similar sounds, the features of a sound's transitions are represented in a convenient form. Utilizing these analysis results, a new sound that inherits the transition features from an entirely different sound will be synthesized.