Scaling of audio signals using frequency domain techniques

Abstract

Scaling of audio signals finds numerous applications in karaoke and personal computer based sound editors. Two types of scaling operations usually performed on audio signals are time scaling and pitch scaling. Time scaling refers to changing the time base of the signal without altering its pitch. The pitch of a signal is the fundamental frequency of vibration of the source of the tone. Pitch scaling refers to dilating the frequency characteristics of the signal while upholding its harmonic relationship and keeping the time base constant. While most of the existing techniques rely on time domain techniques for scaling signals, our approach to the pitch scaling problem has been through frequency domain methods. This results in significant computational saving, as for pitch scaling a signal, there is no need to convert the sampling rate of the time scaled signal. We use a sinusoidal model for representing and modifying audio signals and demonstrate its advantages over existing systems. A new approach for maintaining phase continuity across frames has also been discussed which is computationally more efficient than existing techniques.