HNS: Speech modification based on a harmonic+noise model

HNS (harmonic plus noise synthesis), an analysis/modification/synthesis model based on a harmonic plus noise representation of the speech signal, is presented. Significant improvements over previous work on the subject are proposed at both the analysis and the synthesis stages: a model of harmonically related sinusoids with linearly varying complex amplitudes for the representation of the deterministic part of the signal; a joint time-domain and frequency-domain representations of the stochastic part of the signal; and a pitch-synchronous PSOLA (pitch-synchronized overlap-add)-like synthesis scheme. Informal listening-tests demonstrate the effectiveness of this approach for time-scale modifications.<>