Wake-Word Detection Feedback for Low-Power Dual-Mic Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing audio processing systems face challenges in efficiently detecting wake-words with high accuracy and low power consumption, particularly in always-on voice systems, due to the trade-off between false acceptance and rejection rates, and the inefficiencies in echo cancellation methods using traditional linear filters.
Innovation Solution
Implementing a dual-mic processing system with a low-complexity wake-word model for preliminary screening, followed by advanced verification, and using a displacement-based mathematical model for echo cancellation to accurately detect wake-words while minimizing power consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional linear filters are used for echo cancellation, then device complexity is reduced, but measurement precision of wake-word detection deteriorates
Solution Approach 1:
The system segments wake-word detection into two stages: a low-complexity prescreening stage using simple energy detection, and a high-precision verification stage using advanced acoustic modeling. This segmentation allows the system to achieve high detection accuracy while maintaining low overall complexity by applying complex processing only when necessary.
Solution Approach 2:
The prescreening engine performs preliminary action by quickly identifying potential wake-word candidates using simple energy-based detection before committing resources to more complex verification processing. This preliminary filtering reduces the burden on subsequent verification stages and improves overall system efficiency.
2Speed
If always-on voice processing is implemented, then responsiveness to wake-words is improved, but energy consumption increases
Solution Approach 1:
The system implements periodic action by alternating between low-power prescreening operations and higher-power verification operations. The prescreening engine continuously monitors audio energy at low power consumption, and only triggers verification processing when wake-word candidates are detected, thereby maintaining responsiveness while managing energy usage.
Solution Approach 2:
The system uses self-service by having the prescreening engine automatically filter and prioritize audio segments that warrant further processing. This self-filtering mechanism ensures that energy-intensive verification resources are allocated only to promising candidates, optimizing the balance between detection speed and power consumption.
3Measurement precision
If false acceptance rate is reduced, then detection precision is improved, but false rejection rate increases
Solution Approach 1:
The system segments the detection process into prescreening and verification stages with different threshold characteristics. The prescreening stage uses more sensitive thresholds to capture potential candidates, while the verification stage applies stricter thresholds to confirm actual wake-words. This segmentation allows the system to maintain low false acceptance rates while minimizing false rejections through the two-stage approach.
Solution Approach 2:
The system implements feedback by using verification results to refine prescreening parameters and improve future detections. The interaction between prescreening and verification stages creates a feedback loop that optimizes the balance between false acceptance and false rejection rates over time, improving overall detection reliability.
Data Source
AI summary
An audio processing circuit includes an audio front end (AFE) configured to generate preliminary audio data in response to audio signals received from a plurality of microphones and one or more special purpose engines (SPEs) configured to: determine when the preliminary audio data corresponds to a candidate wake-word; generate prescreening feedback to the AFE in response to the candidate wake-word, wherein the AFE generates, based on the prescreening feedback, targeted audio data; determine when the targeted audio data corresponds to a verified wake-word; and generate verified wake-word data when the targeted audio data corresponds to a verified wake-word.