Acoustic Voice Activity Detection Using Dual-Microphone Noise Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional acoustic voice activity detectors (AVADs) face limitations in accurately detecting speech in noisy environments, leading to false positives and false negatives, which affect the performance of communication devices by either devoicing or failing to suppress noise effectively.
Innovation Solution
The implementation of a Dual Omnidirectional Microphone Array (DOMA) and a combination of acoustic and vibration sensor-based voice activity detection algorithms, which form virtual microphones with distinct noise and speech responses, allowing for improved noise suppression and accurate speech detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional acoustic voice activity detection (AVAD) is used to detect speech in noisy environments, then noise suppression can be achieved, but false positives and false negatives increase, reducing detection accuracy
Solution Approach 1:
The patent segments the speech detection task into two independent detection paths: skin surface microphone (SSM)-based VAD and acoustic AVAD. Each detector operates independently with its own detection logic and thresholds, allowing the system to leverage the strengths of each method while mitigating their individual weaknesses through subsequent combination logic
Solution Approach 2:
The patent merges the outputs of multiple independent voice activity detectors (SSM VAD, acoustic AVAD, and their combinations) using logical operations. The final VAD signal is generated by combining detection results from different modalities (skin contact detection, acoustic analysis) to achieve more reliable speech detection with reduced false positives and negatives
2Measurement precision
If conventional AVAD is tuned to reduce false positives, then denoising performance improves, but false negatives increase, causing devoicing
Solution Approach 1:
The patent applies different detection thresholds and logic rules to different detection scenarios and modalities. The SSM VAD uses one set of detection parameters optimized for contact-based detection, while the acoustic AVAD uses different parameters optimized for acoustic analysis. This allows each detector to operate at its optimal precision level for its specific modality
Solution Approach 2:
The system performs preliminary detection using multiple independent VAD methods before generating the final speech activity determination. By having multiple detectors run in parallel and their results combined through logical operations, the system can identify speech events that might be missed by any single detector, preventing devoicing while maintaining false positive reduction
3Measurement precision
If skin contact is indicated and SSM VAD is used, then false AVAD indications are reduced, but diffuse noises cause false negative indications
Solution Approach 1:
The patent introduces an intermediary combination logic that mediates between the SSM VAD output and the acoustic AVAD output. The final VAD signal is generated by combining these two detection paths using logical operations, allowing the system to use the SSM VAD for accurate speech detection while compensating for its vulnerability to diffuse noises through the acoustic AVAD pathway
4Device complexity
If conventional solutions are used in highly reverberant environments, then device complexity is reduced, but false negative indications increase, leading to devoicing
Solution Approach 1:
The patent implements dynamic selection and combination of detection methods based on environmental conditions and signal characteristics. The system adaptively adjusts which detection paths are active and how their outputs are combined, allowing it to maintain high detection accuracy in challenging environments like highly reverberant conditions while managing device complexity through intelligent resource allocation
Data Source
AI summary
Techniques for acoustic voice activity detection (AVAD) is described, including detecting a signal associated with a subband from a microphone, performing an operation on data associated with the signal, the operation generating a value associated with the subband, and determining whether the value distinguishes the signal from noise by using the value to determine a signal-to-noise ratio and comparing the value to a threshold.