This repository contains the code developed as part of the master thesis "Audio retrieval from big databases by melodic movements: A virtual agent for live coding".
The use of crowdsourced sounds from big databases for creative applications has started to grow rapidly, but real-time exploration of these databases can be a challenging process. In the art of Live Coding, it is a common practice to use prerecorded sounds, whereas MIR and Machine Learning (ML) tools are already being used for helping and expanding the capabilities of the live coding environment. In this work, a theoretical overview about why and how collaborative big datasets and live coding can be combined with the help of MIR and ML algorithms, along with a new system that is implemented for retrieving sounds from Freesound, a big database with broad content, according their melodic movements. With this tool, the user can easily retrieve sounds with a high-level musical characteristic based on the concept of melody, as it is being redefined to fit better the nature of the retrieved sounds and the live coding practices.
Melodic Querying: Search for audio by specifying melodic movements or patterns.
Live Coding Interface: Interact with the virtual agent in real-time, making dynamic queries to the database.
- Install the MelodyLine quark to the Extensions of SuperCollider.
- Run the main_script.py Python file to activate the computational processes that support functionality of the quark.
This project is licensed under the MIT License.