Efficient, high-quality time-scaling of audio signals

Abstract

Enhancements to the classical Waveform Similarity Overlap-Add (WSOLA) algorithm are proposed. As a time-domain approach, it works best for small speed changes and quasi-periodic, monophonic signals. Some of our enhancements are especially effective for small, others for large speed changes. As a consequence, significant improvements for all scaling factors are achieved extending the usability of the new scheme to larger speed changes and more complex signal characteristics. The reduction in computational complexity is analyzed by comparing the number of splice points needed to time-scale the input signal. As will be shown, these are the only points where real signal processing is performed. Therefore, a reduction in their number results in an equivalent decrease in computational demand. Additionally, they are also the only points where artifacts may arise so that, in many cases, a reduction in their number can serve as an indicator for improvements in the signal quality, too.