Audio representations for data compression and compressed domain processing

I certify that I have read this dissertation and that in my opinion it is fully adequate, in scope and in quality, as a dissertation for the degree of Doctor of Philosophy. I certify that I have read this dissertation and that in my opinion it is fully adequate, in scope and in quality, as a dissertation for the degree of Doctor of Philosophy. I certify that I have read this dissertation and that in my opinion it is fully adequate, in scope and in quality, as a dissertation for the degree of Doctor of Philosophy. In the world of digital audio processing, one usually has the choice of performing modiications on the raw audio signal or performing data compression on the audio signal. But, performing modiications on a data compressed audio signal has proved diicult in the past. This thesis provides new representations of audio signals that allow for both very low bit rate audio data compression and high quality compressed domain processing and modiications. In this system, two compressed domain processing algorithms are available: timescale and pitch-scale modiications. Timescale modiications alter the playback speed of audio without changing the pitch. Similarly, pitch-scale modiications alter the pitch of the audio without changing the playback speed. The algorithms presented in this thesis segment the input audio signal into separate sinusoidal, transients, and noise signals. During attack-transient regions of the audio signal, the audio is modeled by transform coding techniques. During the remaining non-transient regions, the audio is modeled by a mixture of multiresolution sinusoidal modeling and noise modeling. Careful phase matching techniques at the time boundaries between the sines and transients allow for seamless transitions between the two representations. By separating the audio into three individual representations, each can be eeciently and perceptually quantized. In addition, by segmenting the audio into transient and non-transient regions, high quality timescale modiications that stretch only the non-transient portions are possible. v vi Acknowledgements First I would like to thank my principal advisor, Prof. Julius O. Smith III. In addition to being a seemingly all-knowing audio guy, our weekly meetings during my last year in school helped me out immensely by keeping me and my research focused and on track. If it were not for the academic freedom he gives me and the other CCRMA grad students, I would not have stumbled across this thesis topic. My next thanks goes out to Tony Verma, …