S.K. Marks and R. Gonzalez (Australia)
Audio and Speech Processing, Time/Frequencyresolution, Sinusoidal Model.
The sinusoidal model is a key component in many speech and audio coding applications, however its reconstruction quality is severely limited by its poor time resolution, and is unable to capture the onset of sinusoidal partials in a signal accurately. This paper proposes two algorithms which improve the onset localisation of sinusoidal tracks. While significant improvement can be achieved by using multiresolution sinusoidal analysis to manage the time/frequency resolution tradeoff in individual frequency-subbands, multiresolution sinusoidal analysis cannot improve upon the inherent time/frequency resolution of the Discrete Fourier Transform. The proposed algorithms work in collaboration with multiresolution sinusoidal analysis to improve the time resolution of the sinusoidal model. This paper will show that these algorithms drastically reduce the average onset error, without adjusting the frequency resolution of the sinusoidal model.
Important Links:
Go Back
