Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2407.02156 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866929406179540992 |
|---|---|
| author | Kroher, Nadine Manangu, Steven Pikrakis, Aggelos |
| author_facet | Kroher, Nadine Manangu, Steven Pikrakis, Aggelos |
| contents | Most contemporary music tagging systems rely on large volumes of annotated data. As an alternative, we investigate the extent to which synthetically generated music excerpts can improve tagging systems when only small annotated collections are available. To this end, we release GTZAN-synth, a synthetic dataset that follows the taxonomy of the well-known GTZAN dataset while being ten times larger in data volume. We first observe that simply adding this synthetic dataset to the training split of GTZAN does not result into performance improvements. We then proceed to investigating domain adaptation, transfer learning and fine-tuning strategies for the task at hand and draw the conclusion that the last two options yield an increase in accuracy. Overall, the proposed approach can be considered as a first guide in a promising field for future research. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2407_02156 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Towards Training Music Taggers on Synthetic Data Kroher, Nadine Manangu, Steven Pikrakis, Aggelos Sound Artificial Intelligence Information Retrieval Machine Learning Audio and Speech Processing I.2 Most contemporary music tagging systems rely on large volumes of annotated data. As an alternative, we investigate the extent to which synthetically generated music excerpts can improve tagging systems when only small annotated collections are available. To this end, we release GTZAN-synth, a synthetic dataset that follows the taxonomy of the well-known GTZAN dataset while being ten times larger in data volume. We first observe that simply adding this synthetic dataset to the training split of GTZAN does not result into performance improvements. We then proceed to investigating domain adaptation, transfer learning and fine-tuning strategies for the task at hand and draw the conclusion that the last two options yield an increase in accuracy. Overall, the proposed approach can be considered as a first guide in a promising field for future research. |
| title | Towards Training Music Taggers on Synthetic Data |
| topic | Sound Artificial Intelligence Information Retrieval Machine Learning Audio and Speech Processing I.2 |
| url | https://arxiv.org/abs/2407.02156 |