Abstract
We propose to use enhanced F0-trajectories, which are extracted using shift-autocorrelation (shift-ACF), for multiple speaker detection in audio monitoring scenarios. After introducing spectral shift-ACF features, their performance in a multiple FO-extraction in the presence of different noise types is estimated for synthetic signal scenarios. Afterwards, at novel method for F0-supertrajectory extraction is proposeds and evaluated for multiple speaker detection in the presence of background noises that typically occur in audio monitoring. It turns out that due to their improved sharpness in representing harmonic components, spectral shift-ACF features outperform classical features in many cases.