Dual-Processor Audio Classifier for Always-On Voice Recognition

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Voice recognition systems and speaker authentication systems face significant processing and power consumption demands, making them challenging for mobile devices, especially when requiring continuous 'always-on' functionality, which can disrupt user experience and drain battery life.

Innovation Solution

An audio classifier is implemented using a dual-processor architecture with hard-wired logic for initial audio activity detection and reconfigurable logic for classification, allowing for efficient classification and activation of voice recognition systems without user intervention, reducing power consumption and processing overheads.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If voice recognition systems operate in 'always-on' mode to improve user experience, then responsiveness and convenience are improved, but power consumption and processing demands increase significantly

Engineering Contradiction:
Improveuser experienceVSAvoidpower consumption
Core Design Contradiction:
Ease of operationVSUse of energy by moving object

Solution Approach 1:

The audio processing system is segmented into two distinct processors: a first processor with hard-wired logic for initial audio activity detection, and a second processor with reconfigurable logic for comprehensive audio classification. This segmentation allows the system to perform basic detection at low power while reserving higher-power processing for when needed, thereby maintaining always-on responsiveness without excessive power consumption.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The first processor performs partial audio analysis by detecting basic audio activity presence, which is sufficient for activation purposes. The second processor performs the remaining classification work only when the first processor detects activity. This partial action approach avoids the excessive power consumption of full always-on processing while maintaining system responsiveness.

Inventive Principle:
Principle #16Partial or excessive action

2Speed

If voice recognition systems continuously monitor audio signals to enable always-on functionality, then system responsiveness is improved, but processing overheads increase

Engineering Contradiction:
Improvesystem responsivenessVSAvoidprocessing overheads
Core Design Contradiction:
SpeedVSDevice complexity

Solution Approach 1:

The processing workload is segmented between two processors with different capabilities. The first processor handles continuous monitoring with simple hard-wired logic for quick audio activity detection, providing fast responsiveness. The second processor handles complex classification only when needed, reducing overall processing overheads while maintaining system responsiveness.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The first processor acts as an intermediary between the continuous audio input and the second processor. It filters and pre-processes audio signals by detecting activity presence, thereby reducing the burden on the second processor and overall system processing overheads while maintaining fast response capability.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Use of energy by moving object

If keyword detectors or button presses are required to activate voice recognition, then power consumption is reduced, but user convenience deteriorates

Engineering Contradiction:
Improvepower consumptionVSAvoiduser convenience
Core Design Contradiction:
Use of energy by moving objectVSEase of operation

Solution Approach 1:

The audio classification system performs self-service by automatically detecting audio activity and classifying signals without requiring user intervention. The dual-processor system autonomously determines when voice commands are present and activates recognition accordingly, eliminating the need for keyword detection or button presses while maintaining low power consumption through selective processing.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP3125244B1Audio classifier
Publication Date: 2019.11.27 NXP BV
  • EP3125244B1 patent drawingFigure 1~2
  • EP3125244B1 patent drawingFigure 3
  • EP3125244B1 patent drawingFigure 4~5

AI summary

The disclosure relates to an audio classifier comprising: a first processor having hard-wired logic configured to receive an audio signal and detect audio activity from the audio signal, and a second processor having reconfgurable logic configured to classify the audio signal as a type of audio signal in response to the first processor detecting audio activity.