Hybrid multiresolution analysis of 'punch' in musical signals

This paper presents a hybrid multi-resolution technique for the extraction and measurement of attributes contained within a musical signal. Decomposing music into simpler percussive, harmonic and noise components is useful when detailed extraction of signal attributes is required. The key parameter of interest in this paper is that of punch. A methodology is explored that decomposes the musical signal using a critically sampled constant-Q filterbank of quadrature mirror filters (QMF) before adaptive windowed short term Fourier transforms (STFT). The proposed hybrid method offers accuracy in both the time and frequency domains. Following the decomposition transform process, attributes are analyzed. It is shown that analysis of these components may yield parameters that would be of use in both mixing/mastering and also audio transcription and retrieval.