Sound Search in User Contributed Databases Using Voice Imitations
Page about my final degree project, containing description and resources.
Link to download
Abstract
Human-Computer Interaction (HCI) research is today a very active source for innovation. Lots of applications and services for web and portable devices require novel ways of interaction to make the experience more entertaining and comfortable. In this project, an improvement to the sound search in a large user-contributed database by using voice imitations is proposed. A vocal query is analyzed, extracting a set of audio features that then are classified with a Support Vector Machine algorithm. The system relies on a study about voice analysis and the user behavior when imitating sounds and how can it be managed in a machine learning workflow. This project includes designing and carrying out user experiments of vocal imitations to have a real exposure to this type of interaction and data.
Resources
- Vocal Imitations from Recording Experiment (Readme)
- Cat Imitations (CSV Descriptors for Weka) (Audio Files)
- Dog Imitations (CSV Descriptors for Weka) (Audio Files)
- Car Imitations (CSV Descriptors for Weka) (Audio Files)
- Drums Imitations (CSV Descriptors for Weka) (Audio Files)
2. Source Data
Video Demo
Supervision
Jordi Janer - jordi . janer at upf . edu
Music Technology Group at the Pompeu Fabra University - Barcelona, Spain
Music Technology Group at the Pompeu Fabra University - Barcelona, Spain