Honselaar, C. and Gort, F. (2002) Timbre segregation and recognition in a model of the auditory system based on associative relaxation oscillator networks. Master's Thesis / Essay, Computing Science.
|
Text
Infor_Ma_2002_CHonselaar_FGort.CV.pdf - Published Version Download (1MB) | Preview |
Abstract
For an animal it is important to be able to recognize meaningful sounds in the auditory scene. Sound recognition is largely based on timbre, a feature uniquely identifying a sound source irrelevant of its pitch. Therefore, the current research presents a model for the auditory system, which performs timbre processing, on the basis of periodic data. A periodicity-segregating network extracts the periodicities of experimental sounds and performs scale normalization in order to achieve pitch independent static audio patterns. After first exploring the capacities for pattern completion and segmentation of a Hebbian learning associative network of relaxation oscillators, it is modified for classification of audio patterns. The associative network appears to successfully segment and complete small patterns, but that does not apply to larger overlapping patterns. In three tests the neural network is trained with two instruments and then classification is performed. Results reveal a classification biased in the direction of one of the instruments. In addition, analysis of the segregated patterns suggests that penodicity alone cannot account for timbre recognition although it might provide some cues.
| Item Type: | Thesis (Master's Thesis / Essay) |
|---|---|
| Degree programme: | Computing Science |
| Thesis type: | Master's Thesis / Essay |
| Language: | English |
| Date Deposited: | 15 Feb 2018 07:30 |
| Last Modified: | 15 Feb 2018 07:30 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/8885 |
Actions (login required)
 |
View Item |