Network Microphone Noise Classification for Private Event Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice-assisted media playback systems face challenges in accurately distinguishing between user commands and environmental noise, leading to false positives and negatives in event detection, while also raising privacy concerns due to the transmission of raw audio data.
Innovation Solution
The system employs network microphone devices equipped with wake-word engines to detect voice inputs, processes sound metadata to classify noise, and uses additional environmental sensors to enhance event detection accuracy, ensuring privacy by transmitting only indecipherable sound metadata.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the system transmits raw audio data for event detection, then event detection accuracy can be improved, but user privacy is compromised
Solution Approach 1:
The patent extracts only the essential acoustic features (spectral centroid, spectral rolloff, spectral flux, zero-crossing rate) from the raw audio signal for transmission and processing, while leaving the raw audio data local to the device. This extraction approach maintains event detection capability while eliminating privacy risks associated with raw audio transmission.
Solution Approach 2:
The system transforms the audio signal from its original raw form into a different parameter space representing acoustic features. By changing the representation from raw waveform to extracted features, the system achieves the same detection functionality while protecting user privacy through mathematical transformation.
2Device complexity
If the system uses only wake-word engines for event detection, then device complexity is reduced, but event detection accuracy deteriorates due to false positives and negatives
Solution Approach 1:
The patent merges multiple detection approaches: wake-word engine detection, noise classification based on acoustic features, and environmental sensor data. By combining these diverse detection mechanisms, the system achieves higher accuracy than any single method alone while maintaining reasonable device complexity through modular integration.
Solution Approach 2:
The system implements a multi-functional detection framework where the same acoustic feature extraction and classification pipeline serves multiple purposes: wake-word detection, noise classification, and environmental event detection. This universal approach improves accuracy across different event types without proportionally increasing complexity.
3Object-affected harmful factors
If the system classifies noise using only sound metadata, then privacy protection is improved, but noise classification accuracy deteriorates
Solution Approach 1:
The patent introduces environmental sensor data as an intermediary that complements sound metadata for noise classification. Sensors such as microphones, accelerometers, and barometers provide additional contextual information about the acoustic environment, enabling accurate noise classification using only processed features rather than raw audio.
Solution Approach 2:
The system creates a composite data structure combining multiple types of processed information: acoustic features from sound metadata, environmental sensor readings, and contextual data. This composite approach achieves accurate noise classification while maintaining privacy protection, as none of the components require raw audio transmission.
Data Source
Figure 1A
Figure 1B
Figure 2A~2B
AI summary
In one aspect, a network microphone device includes a plurality of microphones and is configured to detect sound via the one or more microphones. The network microphone device may capture sound data based on the detected sound in a first buffer, and capture metadata associated with the detected sound in a second buffer. The network microphone device may classify one or more noises in the detected sound and cause the network microphone device to perform an action based on the classification of the respective one or more noises.