Two-Dimensional Fourier Transform Audio Data Structure for Cover Song Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current audio fingerprinting systems struggle to identify cover songs due to variations in key, tempo, instrumentation, and musical structure, often treating them as distinct songs, whereas cover song identification aims to recognize songs with significant musical differences.
Innovation Solution
The use of a data structure based on two-dimensional Fourier transforms (2DFTs) for robust audio representation, which is invariant to key changes, timbral changes, and small tempo deviations, allowing for cross-similarity analysis and distance measurement between audio fingerprints, enabling effective identification of cover songs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If audio fingerprinting systems use traditional matching methods, then exact renditions of reference audio can be identified, but cover versions are treated as different songs and cannot be identified
Solution Approach 1:
The patent transforms the audio representation from traditional time-domain or frequency-domain features to a two-dimensional Fourier transform domain. This dimensional transformation creates a key-invariant representation where cover songs can be matched to reference songs despite key changes, tempo variations, and instrumentation differences. The 2DFT approach adds structural dimensionality that preserves musical composition identity while being invariant to performance variations.
Solution Approach 2:
The system changes the representation parameters of audio data by applying 2DFT to create a transformed domain that is inherently invariant to key changes and timbral variations. This parameter transformation allows the same musical composition to be recognized across different keys, tempos, and instrumentations by operating in a transformed parameter space rather than the original audio space.
2Adaptability or versatility
If audio representation is made robust to key and timbral changes, then cover song identification improves, but computational complexity increases
Solution Approach 1:
The patent pre-computes and stores 2DFT representations of reference songs in a database during an offline preprocessing stage. This preliminary action transforms the computationally intensive 2DFT operation from occurring during real-time query matching to occurring beforehand during database construction. When a cover song query is received, the system only needs to perform the 2DFT on the query audio and compare against pre-transformed references, significantly reducing real-time computational complexity.
Data Source
AI summary
Example systems and methods are audio identification based on data structure are disclosed. An example apparatus includes memory, and one or more processors to execute instructions to execute a constant Q transform on query time slices of query audio, binarize the constant Q transformed query time slices, execute a two-dimensional Fourier transform on query time windows within the binarized and constant Q transformed query time slices to generate two-dimensional Fourier transforms of the query time windows, sequentially order the two-dimensional Fourier transforms in a query data structure, and identify the query audio as a cover rendition of reference audio based on a comparison between the query data structure and a reference data structure associated with the reference audio.