Media Excerpt Extraction via Perceptual Quality Vector Analysis

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current media systems fail to automatically extract the most exciting, representative, or interesting sections of media objects for previews, relying on human input and lacking the ability to generate loops using audio analysis effectively.

Innovation Solution

The method computes perceptual quality vectors for each bar of a media object, sorts distances between bars, and generates a sorted list to identify the most relevant sections, which can be used to create a loop or preview, incorporating acoustic analysis and crossfade techniques for seamless playback.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If fixed ratio extraction is used for media previews, then the extraction process is simple and fast, but the preview quality and representativeness are inadequate

Engineering Contradiction:
Improvepreview qualityVSAvoidextraction process complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces the mechanical fixed-ratio extraction method with an audio analysis-based system that computes perceptual quality vectors, detects beats and bars, and identifies exciting sections through signal processing and pattern recognition algorithms

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

Solution Approach 2:

The system changes from using a single fixed extraction ratio to multiple dynamic parameters including perceptual quality vectors, beat detection results, bar boundaries, and excitement metrics that adapt to the specific characteristics of each media object

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If audio analysis is performed to identify exciting sections, then the preview representativeness is improved, but the processing time and computational complexity increase

Engineering Contradiction:
Improvesection selection accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary audio analysis by computing perceptual quality vectors and detecting beat patterns before final excerpt selection, allowing for more accurate identification of exciting sections without excessive processing time during actual preview generation

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The media object is divided into bars and beats through hierarchical segmentation, allowing the system to analyze and identify exciting sections at multiple temporal resolutions, improving accuracy while managing computational complexity through structured breakdown

Inventive Principle:
Principle #1Segmentation

3Adaptability or versatility

If loops are generated without acoustic analysis, then the generation process is simple, but the ability to capture representative sections is lost

Engineering Contradiction:
Improveloop generation capabilityVSAvoidrepresentativeness
Core Design Contradiction:
Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent replaces simple loop generation with an acoustic analysis-based system that uses perceptual quality vectors, beat detection, and pattern recognition to automatically identify and generate loops from the most exciting and representative sections of media objects

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

Data Source

PatentEP3298513B1Extracting an excerpt from a media object
Publication Date: 2020.01.08 SPOTIFY
  • EP3298513B1 patent drawingFigure 1
  • EP3298513B1 patent drawingFigure 2
  • EP3298513B1 patent drawingFigure 3

AI summary

An excerpt of a media object is extracted by computing, for each bar of an N-bar loop, one or more perceptual quality vectors. For each of the one or more perceptual quality vectors within a search zone (S), one or more distances between bar i and bar i + N is computed and sorted to generate a sorted list of bars.