Music Item Relation Mapping Across Audio and Symbolic Formats
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods struggle to determine relations between music files of different formats and individual tracks or submixes, limiting the ability to relate diverse music files based on various characteristics beyond mere similarity.
Innovation Solution
The use of conditional similarity networks (CSNs) to map music files into multiple subspaces, enabling relations based on characteristics such as complementarity, consecutiveness, mood similarity, and genre, allowing cross-domain relations between audio and symbolic files.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conditional similarity networks are applied to full mixes only, then similarity estimation between songs is improved, but the ability to relate individual tracks and cross-domain files (audio to symbolic) deteriorates
Solution Approach 1:
The patent extends conditional similarity networks to handle multiple music file types (audio and symbolic formats) and granularities (individual tracks and full mixes) through a unified framework. The system processes different input formats through appropriate encoders and maps them to a common embedding space, enabling cross-domain similarity estimation between audio files, symbolic files, tracks, and mixes.
Solution Approach 2:
The patent changes the parameters of the conditional similarity network by introducing format-specific encoders and configurable subspace mappings. The system adjusts the embedding process based on input type (audio/symbolic) and granularity (track/mix), using learnable projection matrices to map different representations into a unified space where similarity can be computed across domains.
2Adaptability or versatility
If the track database grows to include more songs and tracks, then the diversity of music content is improved, but the ability to manually determine relations between tracks deteriorates
Solution Approach 1:
The patent pre-computes embeddings for all tracks and mixes in the database, storing them in a searchable index. When a user queries for related tracks, the system performs efficient similarity searches in the pre-computed embedding space rather than computing similarities on-the-fly, dramatically reducing query time even as the database grows.
Solution Approach 2:
The patent replaces manual track relation determination with automated machine learning models. The conditional similarity networks learn to predict track relationships based on audio and symbolic features, substituting human listening and judgment with automated computational methods that scale to large databases.
3Measurement precision
If conditional similarity networks are used for audio files only, then audio-based similarity is improved, but cross-domain relations (audio to symbolic) deteriorate
Solution Approach 1:
The patent introduces a common embedding space as an intermediary between audio and symbolic domains. Audio files are encoded through audio-specific encoders while symbolic files use symbolic encoders, but both are projected into the same latent space where similarity can be computed. This intermediary space enables cross-domain comparisons by translating different representations into a unified framework.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method of determining relations between music items, the method comprising determining a first input representation for a symbolic representation of a first music item, mapping the first input representation onto to one or more subspaces derived from a vector space using a first model, wherein each subspace models a characteristic of the music items, determining a second input representation for music data representing a second music item, mapping the second input representation onto the one or more subspaces using a second model, determining a distance between the mappings of the first and second input representation in each subspace, wherein the distance represents the degree of relation between the first and second input representation with respect to the characteristic modelled by the subspace.