Audio Stem Identification via Machine Learning Probability Vectors

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods fail to effectively identify which audio stems in a database have been used to create a music track, especially as the database grows, as users cannot listen to all stems to determine their usage.

Innovation Solution

A method using machine learning to train a model on audio stems from multiple songs, predicting stems with a probability of being related to input stems, and employing audio similarity algorithms to find similar stems, along with lead sheet and chord sequence analysis, to identify complementary stems.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If users manually listen to all stems in the database to determine usage, then identification accuracy is improved, but time consumption increases significantly

Engineering Contradiction:
Improvestem identification accuracyVSAvoidtime to identify stems
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent replaces the mechanical/manual listening process with an automated audio analysis system that uses signal processing and machine learning algorithms to identify stems programmatically, eliminating the need for human listeners to manually evaluate each stem

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent introduces an intermediary automated analysis system that acts as a mediator between the stem database and users, automatically determining stem usage through audio feature extraction and comparison without requiring direct user involvement in the evaluation process

Inventive Principle:
Principle #24Intermediary (Mediator)

2Adaptability or versatility

If the stem database grows larger to provide more options, then stem selection versatility is improved, but identification complexity increases

Engineering Contradiction:
Improvestem selection optionsVSAvoididentification system complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent segments the complex identification task into distinct processing stages including audio feature extraction, similarity computation, and ranking, allowing the system to handle large databases efficiently by breaking down the overall complexity into manageable components

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent transforms the identification problem from a complex qualitative assessment into a quantitative parameter-based system using audio features such as spectral characteristics, temporal patterns, and statistical measures, enabling efficient comparison and ranking of stems regardless of database size

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3796305B1Audio stem identification systems and methods
Publication Date: 2022.06.29 SPOTIFY
  • EP3796305B1 patent drawingFigure 1
  • EP3796305B1 patent drawingFigure 2
  • EP3796305B1 patent drawingFigure 3

AI summary

Methods, systems and computer program products are provided for identifying an audio stem. Audio stems (t1, .., tN) are stored on a stem database and songs (S1, ..., SP) made with at least a subset of the plurality of the audio stems (t1, .., tN) are stored on a song database. At least partially composed song (S∗) having a predetermined number of pre-selected stems (k) are received. In turn, a probability vector (or relevance value or ranking) is produced for each stem (t1, .., tN) to be complementary to the at least partially composed song (S∗).