Detection of audio events with repetitive structure using generalized autocorrelations
Abstract
We review several signal transforms for representing repeating structures within audio signals in the timefrequency domain. Based on a recently introduced generalized autocorrelation, the shift-ACF, we demonstrate how multiply repeated audio events may be better represented, hence improving detection performance. Using different examples from audio monitoring, we show how such signal transforms can be applied for audio event detection tasks in realistic scenarios. As a particular example we report on recent evaluations on speech detection in noisy recordings.
