Wake Word Detection Using Contextual Audio Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional wake word detection systems in computing devices often experience high false detection and rejection rates, especially in noisy environments or during ongoing conversations, due to their reliance on acoustic features alone, which can negatively impact user experience.
Innovation Solution
The implementation of a wake word detection system that utilizes a combination of acoustic, environmental, lexical, linguistic, and contextual information, along with probabilistic logic and multiple detectors in series, to make holistic determinations about the presence of a wake word, thereby reducing false detections and rejections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional wake word detection systems rely on acoustic features alone, then the device complexity is low, but the detection accuracy deteriorates with high false detection and rejection rates in noisy environments
Solution Approach 1:
The patent combines multiple detection approaches (acoustic feature detection, keyword spotting, and speech activity detection) into a unified wake word detection system. The classifier integrates features from multiple sources including acoustic features, environmental information, lexical information, linguistic information, and contextual information to make holistic determinations about wake word presence, thereby improving detection accuracy while managing system complexity through integrated architecture
Solution Approach 2:
The detection system uses composite feature sets that combine multiple types of information (acoustic, environmental, lexical, linguistic, and contextual features) similar to how composite materials combine different properties. This multi-faceted approach to feature composition enables the system to achieve higher detection accuracy by leveraging the strengths of different feature types while compensating for their individual weaknesses
2Measurement precision
If the detection system performs comprehensive analysis with multiple features and detectors, then the detection accuracy improves, but the processing latency increases
Solution Approach 1:
The system performs preliminary wake word detection using acoustic features and environmental information before committing to full speech processing. The classifier makes preliminary determinations about wake word presence based on available features, and only proceeds with more computationally intensive processing when the wake word is likely present, thereby reducing overall latency while maintaining accuracy
Solution Approach 2:
The detection system is segmented into multiple independent detectors operating in series (acoustic feature detector, keyword spotter, speech activity detector, and classifier). Each detector processes features independently and contributes to the final decision, allowing for modular optimization where faster detectors can operate in parallel and slower detectors only process when needed, balancing accuracy and latency
3Reliability
If the system uses multiple detectors and comprehensive feature analysis, then false detection rates decrease, but the device complexity increases
Solution Approach 1:
The classifier serves as an intermediary that integrates outputs from multiple detectors (acoustic feature detector, keyword spotter, speech activity detector) and combines them with environmental, lexical, and contextual information. This intermediary component harmonizes the decisions from different detectors and feature sources, reducing false detections through comprehensive analysis while managing complexity through a centralized integration point rather than requiring complex interactions between all components
4Reliability
If the detection system processes more feature types, then the reliability of wake word detection improves, but the computational resources required increase
Solution Approach 1:
The system computes only the necessary subset of features required for reliable wake word detection rather than processing all possible feature types continuously. The classifier selectively processes acoustic, environmental, lexical, linguistic, and contextual features based on their relevance and availability, performing partial computation that achieves sufficient reliability without the excessive energy consumption of comprehensive processing of all possible features
Data Source
AI summary
Features are disclosed for detecting words in audio using contextual information in addition to automatic speech recognition results. A detection model can be generated and used to determine whether a particular word, such as a keyword or “wake word,” has been uttered. The detection model can operate on features derived from an audio signal, contextual information associated with generation of the audio signal, and the like. In some embodiments, the detection model can be customized for particular users or groups of users based usage patterns associated with the users.