UC San Diego
Time scale modication using a sinusoidal model
- Author(s): Daniels, Michelle Lee
- et al.
Time-scale modication of audio signals is a classic digital signal processing problem. While many solutions have been implemented over the years, none work perfectly without audible artifacts for all kinds of input. In most applications, trade-offs must be made between computational efficiency and quality, resulting in different solutions being favored in different contexts and for different classes of input signals. Sinusoidal modeling is an approach which has shown signicant promise for producing high-quality time-scaled audio. Leaving noise and transient signal components for future work in order to focus exclusively on sinusoidal components, this thesis describes a variation on traditional sinusoidal modeling which offers a unique combination of features, including separating the process of identifying individual sinusoidal peaks from the process of combining those peaks into sinusoidal tracks, special treatment of phase during the peak tracking process, and analysis enhancements for sinusoidal tracks with modulating frequency and multichannel input signals