Seminar/Thesis: Transfer learning strategies to classify audio datasets
Background
Machine learning methods can nowadays be very helpful to automatically collect data and build large-sized datasets. However, the retrieved datasets need to be catalogued before being further processed. Classification may require expert knowledge annotation to train the algorithm, and the process may be tedious if done manually. The project aims at implementing a convenient transfer learning technique to automatically train algorithms for the tagging of audio datasets.
Aim
Apply transfer learning techniques to train algorithms automatically to catalogue and tag audio datasets.
Learning objectives
- Analyse audio data from open source repositories
- Apply transfer learning methods to train classification algorithm
- Apply machine learning algorithms to tag audio data automatically
Data
| Project type | Seminar (optional: Master thesis) |
| ECTS | 2.5, 5, 7.5, default: 5 |
| Language | English |
| Period | Winter term 2021-22 |
| Presence time | Virtual seminar, working from remote |
| Useful knowledge | Python, Machine learning |
| Work distribution | 100% algorithm development |
| Med. Eng. designation | Advanced Context Recognition (ACR) |
| StudOn link | Please join |
| First meeting | CDH Seminar and Thesis Introduction, 18th October 16:00 |
| Registration | Via StudOn, obligatory after introduction |
Literature
Up-to-date literature recommendations are provided during the lectures.
Examination
Final presentation and final report.
Contact
Annalisa Baronetto
- Job title: Researcher
- Address: Henkestraße 91, Haus 7, 1. OG
91052 Erlangen
Germany - Phone number: +49 9131 85 23608
- Email: annalisa.baronetto@fau.de