Target Sound Detection With Multi-Stage Low-Power Classification

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Always-on audio context detection systems in electronic devices result in high power consumption, reducing battery life, especially in mobile devices, and increasing system complexity due to the detection of multiple sound events.

Innovation Solution

A multi-stage target sound detector with a first stage using a binary classifier for low-power, always-on operation and a second stage with a more powerful classifier activated only when needed for accurate detection, reducing power consumption while maintaining high-performance sound classification.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If an always-on audio context detection system is used to detect multiple sound events, then the detection capability and accuracy are improved, but the power consumption increases significantly

Engineering Contradiction:

Improvesound event detection capabilityVSAvoidpower consumption

Core Design Contradiction:

ReliabilityVSUse of energy by moving object

Solution Approach 1:

The audio context detection system is divided into multiple stages: a first stage that operates always-on with lower computational complexity to detect potential sound events, and a second stage that activates only when needed to provide detailed analysis. This segmentation allows the system to maintain reliable detection capability while significantly reducing average power consumption by keeping the more intensive processing dormant most of the time.

Inventive Principle:

Principle #1Segmentation

2Adaptability or versatility

If the number of sound events to be detected is increased, then the system versatility is improved, but the system complexity and power consumption increase

Engineering Contradiction:

Improvenumber of detectable sound eventsVSAvoidsystem complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The detection system is segmented into a first stage that handles broad sound event detection and a second stage that handles specific sound event classification. This allows the system to support multiple sound event types without proportionally increasing overall system complexity, as the first stage uses a unified approach that can be adapted to different sound events.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The first stage performs partial detection by identifying potential sound events without requiring full analysis, while the second stage provides complete analysis only when necessary. This partial action approach allows the system to maintain versatility for multiple sound events while reducing the computational complexity required for always-on operation.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS20260031099A1Method and apparatus for target sound detection

Publication Date: 2026.01.29 QUALCOMM INC

AI summary

A device to perform sound detection is disclosed. The device includes a memory including a buffer configured to store audio data. The device also includes one or more processors coupled to the memory. The one or more processors are configured to obtain image data. The one or more processors also are configured to generate, based on the image data, an indication of an environment associated with the audio data. Additionally, the one or more processors are configured to determine, based at least partially on the indication of the environment, whether one or more target sounds detected in the audio data corresponds to a particular set of sound event classes of multiple sets of sound event classes.