Name Detection in ANC Wearables Using Acoustic Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing active noise control (ANC) systems in wearable audio components struggle to effectively switch to conversation mode when a second-party attention seeker tries to get the user's attention, often requiring manual intervention by the user.
Innovation Solution
An automated attention handling system that uses a name embedding model trained via knowledge distillation to detect when a user-enrolled name is invoked in real-time audio samples, automatically triggering the ANC system to switch to conversation mode.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If ANC system suppresses ambient sound to improve audio quality, then noise suppression performance is improved, but user connectivity and awareness of important ambient sounds deteriorates
Solution Approach 1:
The ANC system dynamically adjusts its operating mode based on detected attention-seeking behaviors. When a user-enrolled name is detected in the ambient audio stream, the system transitions from full noise suppression mode to conversation mode, allowing ambient sounds to pass through. This dynamic adaptation resolves the contradiction by maintaining noise suppression during normal audio playback while enabling user connectivity when needed.
Solution Approach 2:
The system changes the noise suppression parameter (ANC filter gain) based on the detected audio context. In normal mode, the ANC filter applies strong suppression to reduce ambient noise. When conversation mode is triggered by name detection, the filter gain is reduced or ANC is temporarily disabled, allowing ambient sounds to be heard. This parameter change enables the system to optimize for different user needs.
2Device complexity
If manual ANC switching is required for conversation mode, then system simplicity is maintained, but user convenience and operation ease deteriorates
Solution Approach 1:
The system automatically detects when conversation mode should be activated by monitoring for user-enrolled names in the ambient audio stream. When a match is found, the system autonomously switches to conversation mode without requiring user intervention. This self-service capability maintains relative system simplicity while dramatically improving ease of operation, as users no longer need to manually disable ANC or press buttons to engage conversations.
3Ease of operation
If automated name detection is implemented, then ease of operation is improved, but device complexity and processing requirements worsen
Solution Approach 1:
The audio processing pipeline is segmented into distinct functional stages: ambient audio capture, name detection using embedded models, attention-seeking behavior identification, and ANC mode switching. This segmentation allows each component to be optimized independently and reduces overall processing complexity by dividing the complex task of automated attention handling into manageable modules that can process information sequentially.
Solution Approach 2:
The patent uses acoustic segments as an intermediary representation between raw audio input and the final ANC control decision. The name embedding model converts audio samples into acoustic segments that can be compared against enrolled names. This intermediary representation simplifies the matching process and reduces computational complexity compared to direct audio pattern recognition, while still enabling automated detection.
Data Source
AI summary
Automated attention handling techniques are described herein for use with wearable audio components with active noise control (ANC) to suppress ambient sound. A name embedding model is trained automatically to convert name audio samples into acoustic segments based on a knowledge distillation model. The name embedding model is used to generate reference embeddings for each of a user-enrolled set of names, and a relation network and a false rejection network are also trained. In real-time operation, the name embedding model converts real-time audio samples to real-time embeddings, the relation network compared the real-time embeddings to the reference embeddings to look for candidate matches, and the false rejection network validates the candidate matches to detect when one of the user-enrolled names has been invoked. Detecting such an invocation automatically triggers the ANC to switch to a conversation mode.