A constant-Q spectral transformation with improved frequency response

This paper introduces a new transform intended for audio processing. The proposed transform exhibits two interesting features in the frequency domain, namely: a constant-Q characteristic and a steep response for each output channel. Constant-Q implies that the spectral description of the transformed signal is performed along a log-like scale, as opposed to the linear scale of standard transforms, such as the discrete Fourier transform (DFT). The consequent variable resolution makes the new transform especially suited for the analysis of musical audio signals. The improved frequency response, as compared also to the DFT, is achieved by its implementation as a bank of very selective filters based on the frequency-response masking (FRM) approach. Such selectivity results in higher isolation between adjacent channels of the overall transform. Application of the new transform to the analysis of musical signals is illustrated through a computer experiment.