Solo Voice Detection Via Optimal Cancellation

Abstract

Automatically identifying sections of solo voices or instruments within a large corpus of music recordings would be useful e.g. to construct a library of isolated instruments to train signal models. We consider several ways to identify these sections, including a baseline classifier trained on conventional speech features. Our best results, achieving frame level precision and recall of around 70%, come from an approach that attempts to track the local periodicity of an assumed solo musical voice, then classifies the segment as a genuine solo or not on the basis of what proportion of the energy can be canceled by a comb filter constructed to remove just that periodicity. This optimal cancelation filter has other applications in pitch tracking and separating periodic and aperiodic energy.

Detecting Solo Phrases in Music Using Spectral and Pitch-related Descriptors

“…Although the initial observations were promising, the overall results did not satisfy the prospects of the research: the derived decision function was far too noisy to discriminate between solo and non solo parts and resulted in a percentage of 56% correctly assigned frames. Pitch detection via autocorrelation together with a cancellation filter was used by Smit and Ellis (2007) in order to find single voice phrases in baroque music. The output of the autocorrelation was fed into a comb filter to cancel all harmonic parts of the analysed signal excerpt.…”