On the Use of Memory Models in Audio Features
Karl Kristoffer Jensen · 2011
Audio feature estimation is potentially improved by including higher- level models. One such model is the Short Term Memory (STM) model. A new paradigm of audio feature estimation is obtained by adding the influence of notes in the STM. These notes are identified when the perceptual spectral flux has a peak, and the spectral content that is increased by the new note is added to the STM. The STM is exponentially fading with time span and number of elements, and each note only belongs to the STM for a limited time. Initial experiment regarding the behavior of the STM shows promising results, and an initial experiment with sensory dissonance has been undertaken with good results.