Audio Activity Detection Using Temporal Context and Static Weights

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing Signal Activity Detection (SAD) algorithms face challenges in rapidly changing noise environments, requiring significant preamble time for adaptation and manual tuning, leading to unreliable decisions and increased complexity.

Innovation Solution

An apparatus and method using a feature extractor, temporal context extractor, and feature combiner with static combination weights in an open loop design, allowing for automatic global optimization and fast adaptation to noise conditions, independent of past decisions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If adaptive long-term noise estimation techniques are used, then noise estimation accuracy is improved, but device complexity and preamble time increase

Engineering Contradiction:
Improvenoise estimation accuracyVSAvoidalgorithm complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts only the essential short-term statistical features (minimum power levels) needed for noise estimation, discarding the complex long-term adaptive mechanisms. This extraction approach maintains adequate noise estimation accuracy while significantly reducing algorithm complexity and eliminating the need for lengthy preamble periods.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent segments the noise estimation process into independent short-term analysis windows, evaluating minimum power levels in each window without requiring continuous long-term adaptation. This segmentation allows the system to achieve reliable noise estimates quickly without the computational burden of long-term statistics.

Inventive Principle:
Principle #1Segmentation

2Measurement precision

If adaptive long-term noise estimation techniques are used, then noise estimation accuracy is improved, but response time to changing noise environments worsens

Engineering Contradiction:
Improvenoise estimation accuracyVSAvoidresponse time to noise changes
Core Design Contradiction:
Measurement precisionVSSpeed

Solution Approach 1:

The patent implements a dynamic short-term statistical analysis that continuously updates noise estimates based on recent signal characteristics. This dynamic approach allows the system to adapt quickly to changing noise environments while maintaining adequate estimation accuracy, unlike static long-term adaptive methods that suffer from inherent latency.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent performs preliminary noise estimation using short-term minimum power level measurements before making VAD decisions. This preliminary action enables the system to have noise estimates ready immediately without waiting for long-term adaptation periods, thus improving response time while maintaining sufficient accuracy.

Inventive Principle:
Principle #10Preliminary action

3Device complexity

If minimum statistics approach is used, then device complexity is reduced, but measurement precision in complex acoustic environments worsens

Engineering Contradiction:
Improvealgorithm complexityVSAvoidsignal activity detection accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent merges the simple minimum statistics approach with additional acoustic features (spectral shape, zero-crossing rate, energy measures) to compensate for the limitations of minimum statistics alone. This combination maintains low device complexity while significantly improving signal activity detection accuracy in complex acoustic environments.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent creates a composite detection approach by combining multiple feature types (power level statistics, spectral characteristics, temporal features) into a unified VAD decision framework. This composite method achieves high detection accuracy in complex environments while keeping individual component complexities low.

Inventive Principle:
Principle #40Composite materials

4Adaptability or versatility

If manual tuning is used for VAD optimization, then adaptability to specific applications is improved, but ease of operation and global optimization worsen

Engineering Contradiction:
Improveapplication-specific optimizationVSAvoidautomatic optimization
Core Design Contradiction:
Adaptability or versatilityVSEase of operation

Solution Approach 1:

The patent implements self-service through automated feature extraction and combination rules that work universally across different applications without requiring manual tuning. The system automatically adapts to various acoustic environments and application requirements through its inherent short-term statistical analysis and multi-feature combination approach, eliminating the need for manual optimization while maintaining high versatility.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP4648049A1Apparatus and method for processing a pre-processed audio input signal to obtain an activity output signal
Publication Date: 2025.11.12 FRAUNHOFER GESELLSCHAFT ZUR FORDERUNG DER ANGEWANDTEN FORSCHUNG EV
  • EP4648049A1 patent drawingFigure 1
  • EP4648049A1 patent drawingFigure 2
  • EP4648049A1 patent drawingFigure 3

AI summary

Disclosed is an apparatus for processing a pre-processed audio input signal to obtain an activity output signal; comprising a feature extractor to receive the pre-processed audio input signal comprising a spectral energy and/or a noise-free spectral energy in at least one frequency band of a plurality of frequency bands. The feature extractor extracts at least one feature depending on the spectral energy and/or depending on the noise-free spectral energy. A temporal context extractor receives the at least one extracted feature and extracts at least one temporal context feature based on the at least one extracted feature, or receives at least one further extracted feature and extracts at least one temporal context feature based on the at least one further extracted feature. A feature combiner receives and combines the features using the plurality of static combination weights to obtain the output signal.