Files and links (1)
PDFOpen Access (License Unspecified), Open Access
Dissertation
Employing sparsity in the joint estimation of sound source and acoustic channel parameters for natural sounds
Doctor of Philosophy (Ph.D.), Drexel University
May 2011
DOI:
https://doi.org/10.17918/etd-3514
Abstract
The human ear experiences audio on a continual basis, most often in a modified form of the original source signal. Such modification is largely due to the acoustic and physical properties of the environment, which alter the original, or "true," signal propagated by the sound source by changing the direction and energy of the direct and reflected paths of the sound before it reaches the ear. As a result, multiple scaled and delayed copies of the original signal are received. The sum of thereceived paths is characterized by the acoustic channel. Assuming that the modification of the sound source signal is a linear process, the received audio can be thought of as a convolution of the sound source signal and acoustic channel. In various audio signal processing tasks, including recognition and identification, manipulation of the true source by the acoustic channel can be detrimental when attempting to generate precise statistical models, ultimately leading to lower performance. Previous work in the field of dereverberation (removing the effects of the acoustic channel) has most often focused on inverse filtering employing at least two audio channels or observations. Generally, prior algorithms have attempted to estimate only the channel and not the original sound source signal. Such algorithms consist of two knowns (observed audio recordings of the same source) and two unknowns (acoustic channels). In this research, model parameters for both the original sound source signal and acoustic channel are estimated simultaneously using only a single observation channel. Expanding upon prior work, the additional constraint of sparsity is enforced, causing the model parameters to have a small number of nonzero terms. This approach aims to provide accurate estimates of the original source and acoustic channel while removing the channel effects from the observation and reducing noise levels of various natural sounds. These benefits are ideal for applications such as music transcription from live or reverberant sound data and speaker identification systems in diverse acoustic environments. In addition, the encoded audio signals may assist in recognition tasks and in characterization of natural sounds by revealing the underlying signalstructure.
Metrics
20 File views/ downloads
24 Record Views
Details
- Title
- Employing sparsity in the joint estimation of sound source and acoustic channel parameters for natural sounds
- Creators
- Travis Michael Doll - DU
- Contributors
- Youngmoo Kim (Advisor) - Drexel University (1970-)
- Awarding Institution
- Drexel University
- Degree Awarded
- Doctor of Philosophy (Ph.D.)
- Publisher
- Drexel University; Philadelphia, Pennsylvania
- Resource Type
- Dissertation
- Language
- English
- Academic Unit
- College of Engineering (1970-2026); Electrical (and Computer) Engineering [Historical]; Drexel University
- Other Identifier
- 3514; 991014632279404721