Shared Audio Feature Extraction for Always-On Multi-Task Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing audio analysis systems on devices like smartphones face inefficiencies due to redundant audio processing operations across different tasks, leading to increased computational load, memory consumption, and power consumption, especially in always-on applications.
Innovation Solution
Integrating and sharing audio processing algorithms such as keyword spotting, acoustic event detection, and acoustic scene classification by combining low-level feature extraction and selectively sharing high-level feature extraction, reducing redundant operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If separate audio processing operations are used for each audio task, then each task can be performed independently with dedicated hardware, but computational load, memory consumption, and power usage increase unnecessarily
Solution Approach 1:
The patent implements a shared audio processing pipeline where a single set of audio processing operations (including feature extraction, neural network processing, and classification) serves multiple audio tasks simultaneously. The system can perform keyword spotting, acoustic event detection, and acoustic scene classification using the same hardware resources, making the system multi-functional without requiring separate dedicated hardware for each task.
Solution Approach 2:
The patent merges previously separate audio processing operations into a unified pipeline. The audio input is processed through common feature extraction and neural network layers that are shared across different tasks, combining multiple functions into a single integrated system that reduces overall computational overhead.
2Adaptability or versatility
If separate audio processing operations are used for each audio task, then each task has its own fixed function hardware, but computational load and memory consumption increase
Solution Approach 1:
The system achieves versatility through a universal audio processing pipeline that can be configured for different tasks. The same hardware infrastructure supports multiple audio analysis functions (keyword spotting, event detection, scene classification) by processing audio through shared feature extraction and neural network operations, eliminating the need for separate fixed-function hardware for each task.
Solution Approach 2:
The patent segments the audio processing system into modular components: a shared front-end feature extraction module, shared neural network processing layers, and task-specific output interpretation modules. This segmentation allows the system to maintain complexity only where necessary (in the shared processing core) while keeping task-specific adaptations minimal and manageable.
3Productivity
If separate audio processing operations are used for each audio task, then each task can be always-on, but battery usage increases due to redundant operations
Solution Approach 1:
The patent combines multiple always-on audio tasks into a single processing pipeline where the computational work is performed once and shared across all tasks. The system processes audio through common neural network operations simultaneously for keyword spotting, event detection, and scene classification, eliminating redundant computations and significantly reducing the energy cost of maintaining multiple always-on functions.
Solution Approach 2:
The unified audio processing system provides multi-functional capability that allows all audio tasks to remain always-on without proportionally increasing power consumption. The shared processing operations serve multiple purposes simultaneously, enabling the system to maintain high productivity across all audio analysis functions while keeping battery consumption manageable through efficient resource utilization.
Data Source
AI summary
A method, system, and article of multiple task audio analysis uses shared audio processing operations.