Sample Search: Audio Sample Search and Exploration
Main Contributors: Wonil Kim
The Sample Search project focuses on the task of rapidly and accurately searching for specific samples in a large music audio sample database as per user requests. This research aims to develop new methodologies for effectively classifying and searching various types of music samples, especially individual and loop samples.
- Similar Sample Search: Develop algorithms to identify and search for audio samples similar to a criterion sample provided by the user. This will quantify the similarity between audio samples and match corresponding samples within the database.
- Conditional Sample Search: Develop advanced features to search and filter samples based on specific conditions set by the user, such as timbre, rhythm, emotion, genre, etc. Users can also find samples more precisely matching their desired characteristics through specific conditions like 'sharper', 'closer to a jazz style', and so on.
- Reference Track Sample Search: Implement a feature that identifies and searches for samples included in a reference track provided by the user. This enables users to gain inspiration from existing musical works or to find the source of specific sounds for integration into their own creations.
- Sample Embedding Visualization: Visualize a large amount of sample data to enable users to easily understand the relationships and distribution among samples. Particularly, by constructing embeddings in a manner similar to human cognitive systems, it assists users in performing searches that align with their intuition.
- Sample Modification and Generation: Extend the functionality beyond mere searching to include the modification and generation of new samples. This feature allows users to linearly fill the gaps between the searched samples, creating a continuum of sounds. It involves transforming existing samples or synthesizing new ones to achieve a seamless blend or transition among selected samples, catering to creative needs like producing novel sounds or fine-tuning existing ones to better fit a musical piece.
Ultimate Goal and Vision
The ultimate aim of the Sample Search project is to revolutionize the way artists interact with music samples, by integrating diverse functionalities into a unified application. This application will enable users to effortlessly search for, modify, and generate samples, aligning with their creative vision. This approach aims to empower users to lay a solid foundation for their musical compositions, leveraging state-of-the-art technology in sample handling, and fostering a new wave of creativity in the field of music production.
Related Publications
-
Drum Sample Retrieval from Mixed Audio via a Joint Embedding Space of Mixed and Single Audio Samples
Wonil Kim and Juhan Nam
Proceedings of the 149th Audio Engineering Society Convention (AES), 2020 [paper]