Neural Net Processor Architecture for Low-Power Audio Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional audio processing systems consume excessive electrical power, leading to reduced performance and battery discharge in portable devices, as they require continuous operation of high-power main system processors for audio recognition tasks.
Innovation Solution
A neural net processor architecture is introduced, which is designed to perform audio recognition with minimal power consumption by integrating specialized hardware for digital signal processing, allowing the main system processor to remain in a low-power state until necessary, and utilizing a graph memory and hardware accelerators for efficient computations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional audio processing systems continuously operate main system processors for audio recognition, then audio recognition performance is maintained, but electrical power consumption increases significantly
Solution Approach 1:
The patent segments the audio processing system into two distinct parts: a low-power dedicated audio processing unit that handles continuous audio recognition tasks, and a main system processor that operates intermittently. This segmentation allows the main processor to remain in sleep mode during normal operation, significantly reducing power consumption while maintaining audio recognition capability through the dedicated unit.
Solution Approach 2:
The patent introduces a dedicated audio processing unit as an intermediary between the microphone and the main system processor. This intermediary handles the power-intensive audio recognition tasks independently, preventing the main processor from needing to continuously operate, thus resolving the contradiction between maintaining performance and reducing power consumption.
2Reliability
If main system processor operates continuously for audio processing, then processing quality is maintained, but battery life decreases
Solution Approach 1:
By segmenting processing responsibilities between a dedicated audio unit and the main processor, the system maintains high processing quality through specialized hardware while extending battery life by keeping the power-hungry main processor in low-power states during normal audio recognition operations.
3Use of energy by moving object
If power budget is decreased to meet low power requirements, then power consumption is reduced, but processing speed and quality suffer
Solution Approach 1:
The patent applies local quality by creating a specialized audio processing unit with dedicated hardware optimized specifically for audio recognition tasks. This localized optimization maintains high processing speed and quality for audio tasks while the rest of the system operates at lower power levels, resolving the contradiction between power consumption and processing performance.
Data Source
AI summary
A system for operating a headphone can include a primary processor to control the headphone and operate in a low-power state, a cup portion having a microphone to receive an input, a listening sub-system to convert the input into an output signal, and a neural net processor to receive the output signal from the listening sub-system and determine whether to generate a wake signal based on the received output signal.