Vaghy, Julia (2022) Automatic Drum Transcription Using Template-Initialized Variants of Non-negative Matrix Factorization. Bachelor's Thesis, Artificial Intelligence.
|
Text
bAI_2022_VaghyJ.pdf Download (1MB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
toestemming.pdf Restricted to Registered users only Download (123kB) |
Abstract
Low-quality acoustic drum datasets limit the applicability of automatic drum transcription (ADT) systems; however, this may be circumvented by template initialization in cases where the drummer can provide sound samples of their specific drum kit. To explore performance in these scenarios, the present project assesses the accuracy of different template-initialized non-negative matrix factorization (NMF) variants on the ADT task in the presence of background noise. Performance is evaluated for NMFD, the deconvolutional variant with 2-dimensional spectrotemporal templates, and NMF, the original variant with 1-dimensional spectral templates. Three template adaptivity conditions are considered: adaptive, semi-adaptive, and fixed. Furthermore, the effect of additional noise template components is explored. Performance is evaluated on a synthetic dataset containing 20 drum loops and corresponding instrument samples, each on four noise conditions: none, mild, loud, and extreme. Initialization templates are derived from the drum loop's respective sample sounds on the given noise condition. The results suggest that adaptive NMFD is best suited for the task with F = 0.83 ± 0.13, 0.83 ± 0.1, 0.74 ± 0.11, and 0.69 ± 0.14, on the four noise conditions, respectively. Additional noise template components do not lead to significant performance improvement in either of the conditions.
| Item Type: | Thesis (Bachelor's Thesis) |
|---|---|
| Supervisor name: | Lawrence, C.P. and Jaeger, H. |
| Degree programme: | Artificial Intelligence |
| Thesis type: | Bachelor's Thesis |
| Language: | English |
| Date Deposited: | 18 Aug 2022 11:42 |
| Last Modified: | 18 Aug 2022 11:42 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/28429 |
Actions (login required)
 |
View Item |