Two separate methods are presented to aid in the creation and evaluation of acoustic simulations. The first is a hybrid method that allows separate low and high-frequency acoustic responses to be combined into a single broadband response suitable for auralization. The process consists of four steps: 1) creating separate low-frequency and high-frequency responses of the system of interest, 2) interpolating between the two responses to get a single broadband magnitude response, 3) adding amplitude modulation to the high-frequency portion of the response, and 4) calculating approximate phase information. An experimental setup is used to validate the hybrid method. Listening tests are conducted to assess the realism of simulated auralizations compared to measurements. The listening tests confirm that the method is able to produce realistic auralizations, subject to a few limitations. The second method presented is a machine learning approach for predicting human perceptions of audio fidelity and sound quality. Several algorithms are compared and various audio features considered in developing the machine learning models. The developed models accurately predict human perceptions of audio fidelity and sound quality in three distinct applications: assessing the fidelity of compressed audio, evaluating the fidelity of simulated audio, and comparing the sound quality of loudspeakers. The high accuracies achieved confirm that machine learning models could potentially supplant listening tests, significantly decreasing the time required to assess audio quality or fidelity.
| Identifer | oai:union.ndltd.org:BGMYU2/oai:scholarsarchive.byu.edu:etd-9955 |
| Date | 08 April 2021 |
| Creators | Miller, Andrew Jared |
| Publisher | BYU ScholarsArchive |
| Source Sets | Brigham Young University |
| Detected Language | English |
| Type | text |
| Format | application/pdf |
| Source | Theses and Dissertations |
| Rights | https://lib.byu.edu/about/copyright/ |
Page generated in 0.0018 seconds