UC San Diego

This dissertation presents two extensions to the phase vocoder method of sound analysis and synthesis, as well as an examination of the author's spectral subtraction audio works based on the phase vocoder. The phase vocoder technique has proved to be an effective method of "time stretching" musical sounds, however, some musical features of the original sounds are not maintained during the transformation. We consider here maintaining two of these aspects: noise levels, and sub-audio (vibrato and tremolo) characteristics within the phase vocoder. The noise levels are maintained by determining the "sinusoidality" of each spectral component, separating the spectral energy into sinusoidal and noise energy based on these sinusoidality measurements, and modulating the noise based spectrum before re-synthesizing with the traditional IFFT method. The vibrato and tremolo (sub-audio modulation) rates of a sound are maintained by modeling each channel of the phase vocoder in a higher order spectral domain, removing the modulations in this domain, and re-imposing them after time-dilation. Finally, several of the author's audio works involve use of the phase vocoder representation of sound to subtract spectral components from a long term average of a time varying sound. These works will be considered along with their aesthetic motivations and technical implementations