Granted Patent B1

US 12,608,419 · App. 19/189,941 · Granted Apr 21, 2026

Inventors: Geert Trooskens (Meise, BE); Lamara De Brouwer (Malibu, CA); Benjamin Ray Resnick (Plano, TX); Walter Adolf De Brouwer (Malibu, CA)

Assignee: Sound Patrol Inc.

G06F16/632G06F16/683

Abstract

Claims (54)

1 . A system for detecting musical infringement in symbolic music, the system comprising:

one or more processors; and

one or more memories configured to store instructions that, when executed by the one or more processors, perform operations comprising:

converting, using a transcription model, an audio input into a symbolic format when a first musical composition is provided as audio, wherein the transcription model is trained to extract a main melody;

generating a set of k-mer sequences from the first musical composition, wherein each k-mer sequence comprises a plurality of consecutive notes;

indexing the k-mer sequences in a data structure that enables retrieval, wherein the data structure is configured to handle dynamic conditioning vectors for selectively focusing on melody, rhythm, key, insertions, and deletions without requiring full reindexing;

comparing the k-mer sequences of the first musical composition to a plurality of k-mer sequences associated with a plurality of reference musical compositions stored in one or more databases;

estimating a similarity measure between the first musical composition and at least one reference musical composition in the plurality of reference musical compositions;

in response to determining that the similarity measure exceeds a threshold, performing a refined alignment of the first musical composition and the at least one reference musical composition using a local sequence alignment function adapted for music, wherein the local sequence alignment function accounts for key shifts, rests, and variations in melodic or rhythmic patterns;

determining, based on the refined alignment, whether the first musical composition includes at least one musical fragment that infringes upon or regurgitates a portion of the at least one reference musical composition; and

based on the determination, detecting music infringement by identifying that the first musical composition was recreated from copyrighted works, thus ensuring that content owners can detect unauthorized uses of their material.

2 . The system of claim 1 , wherein at least one data cleaning step is applied to the first musical composition prior to generating the plurality of k-mer sequences, and wherein the at least one data cleaning step comprises filtering out notes with anomalous pitches, converting chords to consecutive notes, normalizing keys, and removing sequences that fall below a minimal entropy threshold.

3 . The system of claim 1 , wherein the local sequence alignment function is performed by representing the first musical composition and the at least one reference musical composition on a quantized temporal grid, constructing a differential similarity matrix based on note comparisons and rest alignments.

4 . The system of claim 1 , wherein a larger set of conditioning vectors is applied upon determining that an initial similarity measure is below a similarity threshold.

5 . The system of claim 1 , wherein the audio input is received and converted into the symbolic format using the transcription model trained to isolate a main melody.

6 . The system of claim 1 , wherein the plurality of k-mer sequences is indexed by storing references to each k-mer in a hash table optimized for retrieval and associating each k-mer with metadata including instrument identifiers, positional data, and song identifiers.

7 . The system of claim 1 , wherein a first MinHash sketch is generated for the first musical composition and a second MinHash sketch is generated for the at least one reference musical composition, and wherein each MinHash sketch represents a reduced-size approximation of the set of k-mers.

8 . The system of claim 1 , wherein scoring elements are included in the local sequence alignment function such that matched notes yield a positive score, rest-to-rest matches yield a neutral score, and mismatches or unmatched notes incur penalties, thereby differentiating subtle variations from direct copying.

9 . The system of claim 1 , wherein results obtained from audio fingerprinting analysis are overlayed on spectrogram data with the local sequence alignment function results for combining symbolic sequence alignment with audio-based features to detect subtle variations, remixes, or regurgitated AI-generated content that are not captured by symbolic analysis alone.

10 . The system of claim 1 , wherein network representations of the first musical composition and the at least one reference musical composition are constructed, and wherein nodes represent notes and directed edges represent transitions between notes, and compare the network representations to identify recurring motifs or progressions that may indicate regurgitated content.

11 . A method for detecting musical infringement in symbolic music, the method comprising:

receiving a first musical composition in a symbolic format that encodes musical elements;

generating a set of k-mer sequences from the first musical composition, each k-mer sequence comprising a series of notes;

indexing the set of k-mer sequences using a data structure configured to dynamically adjust query conditions for melody-only, rhythm-only, or combined melody-rhythm matches;

comparing the k-mer sequences of the first musical composition to a plurality of k-mer sequences associated with a plurality of reference musical compositions stored in one or more databases;

estimating a similarity measure between the first musical composition and at least one reference musical composition in the plurality of reference musical compositions;

determining that the at least one reference musical composition has a similarity measure exceeding a similarity threshold;

performing a local sequence alignment between the first musical composition and the at least one reference musical composition to account for transpositions, rest notes, and partial matches;

determining whether the first musical composition includes content that infringes upon the at least one reference musical composition based on the local sequence alignment; and

based on the determination, detecting music infringement by identifying that the first musical composition was recreated from copyrighted works, thus ensuring that content owners can detect unauthorized uses of their material.

12 . The method of claim 11 , further comprising applying one or more data cleaning steps to the first musical composition prior to generating the plurality of k-mer sequences, wherein the data cleaning steps include filtering out notes with anomalous pitches, converting chords to consecutive notes, normalizing keys, and removing sequences that fall below a minimal entropy threshold.

13 . The method of claim 11 , wherein performing the local sequence alignment comprises:

representing both the first musical composition and the at least one reference musical composition on a quantized temporal grid;

constructing a differential similarity matrix based on note comparisons and rest alignments; and

utilizing a dynamic programming algorithm selected from the group consisting of Smith-Waterman and Needleman-Wunsch to identify locally optimal alignments indicating potential infringements.

14 . The method of claim 11 , further comprising:

upon determining that an initial similarity measure is below the similarity threshold, applying a larger set of conditioning vectors to broaden a search criteria.

15 . The method of claim 11 , further comprising converting an audio input into the symbolic format using a transcription model, wherein the transcription model is trained to isolate a main melody, and wherein the main melody is used for generating the plurality of k-mer sequences and subsequent similarity estimation.

16 . The method of claim 11 , wherein indexing the plurality of k-mer sequences further comprises:

storing references to each k-mer in a hash table optimized for retrieval, and

associating each k-mer with metadata including an instrument identifier, positional data, and a song identifier.

17 . The method of claim 11 , further comprising generating a first MinHash sketch for the first musical composition and a second MinHash sketch for the at least one reference musical composition, wherein each MinHash sketch represents a reduced-size approximation of the set of k-mer sequences.

18 . The method of claim 11 , wherein the local sequence alignment includes scoring elements such that matching notes have a positive score, rest-to-rest matches have a neutral score, and mismatches or unmatched notes incur a penalty.

19 . The method of claim 11 , further comprising overlaying results from an audio fingerprinting module that analyzes spectrograms of the first musical composition and the at least one reference musical composition, thereby combining symbolic alignment with audio-based features to detect subtle variations in tempo, instrumentation, or style that may still constitute infringement.

20 . One or more non-transitory, computer-readable media storing instructions thereon that cause one or more processors to perform operations comprising:

receiving a first musical composition in a symbolic format that encodes musical elements;

generating a set of k-mer sequences from the first musical composition, each k-mer sequence comprising a series of notes;

indexing the set of k-mer sequences using a data structure configured to dynamically adjust query conditions for melody-only, rhythm-only, or combined melody-rhythm matches;

comparing the k-mer sequences of the first musical composition to a plurality of k-mer sequences associated with a plurality of reference musical compositions stored in one or more databases;

estimating a similarity measure between the first musical composition and at least one reference musical composition in the plurality of reference musical compositions;

determining that the at least one reference musical composition has a similarity measure exceeding a similarity threshold;

performing a local sequence alignment between the first musical composition and the at least one reference musical composition to account for transpositions, rest notes, and partial matches;

determining whether the first musical composition includes content that infringes upon the at least one reference musical composition based on the local sequence alignment; and

based on the determination, detecting music infringement by identifying that the first musical composition was recreated from copyrighted works, thus ensuring that content owners can detect unauthorized uses of their material.

Assignments (1)

Continuity (1)

References Cited (22)