Voice Activity Detector Combining Multiple VADs for Noise Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice activity detectors (VADs) face challenges in accurately distinguishing speech from non-stationary background noise, particularly babble noise, leading to increased activity and reduced system capacity, especially at low signal-to-noise ratios (SNR), and often result in speech clipping and reduced DTX efficiency.
Innovation Solution
The solution involves combining the outputs of multiple VADs using logical AND or OR operations to generate a modified primary decision, allowing an aggressive external VAD to override the primary VAD in specific conditions, thereby reducing excessive activity and improving detection accuracy without affecting normal operation at good SNR conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single primary VAD is used for speech detection, then the device complexity is low, but the measurement precision of speech activity detection deteriorates in noisy environments
Solution Approach 1:
The system segments the speech detection task into multiple independent VAD instances, each analyzing the signal with different parameters or algorithms. The primary VAD and external VAD operate independently and their decisions are combined through logical operations, allowing each detector to specialize in different aspects of speech detection without increasing individual complexity
Solution Approach 2:
The system merges the output decisions of multiple VAD instances through logical AND or OR operations to form a final speech activity decision. This combination approach integrates the strengths of different detectors while maintaining relatively simple combination logic, resolving the contradiction between using multiple detectors and system complexity
2Reliability
If the primary VAD operates aggressively to detect all speech, then the speech detection coverage is improved, but false activity on background noise increases
Solution Approach 1:
The system introduces an external VAD as an intermediary decision-making layer that processes the same input signal independently. Its output serves as a filter or validator for the primary VAD's decisions, reducing false positives from aggressive detection while maintaining speech coverage through the combination logic
Solution Approach 2:
The combination of multiple VAD decisions creates a feedback mechanism where each detector's output influences the final decision. The logical combination operations provide implicit feedback that suppresses false alarms from any single aggressive detector while preserving true speech detections across multiple detectors
3Measurement precision
If multiple external VADs are combined to improve detection accuracy, then the measurement precision is improved, but the device complexity increases
Solution Approach 1:
The system segments the detection function across multiple independent VAD instances that can be implemented as separate modules or processes. Each VAD handles a specific detection task independently, and their outputs are combined through simple logical operations, making the complexity manageable through functional segmentation
Solution Approach 2:
The system employs multiple VAD instances that can use the same base algorithm but with different parameter settings or training data, making each detector universal in its approach while specialized in different conditions. This allows reuse of proven detection algorithms across multiple instances without proportionally increasing implementation complexity
4Reliability
If the primary VAD uses high sensitivity threshold, then the speech detection coverage is improved, but the productivity of DTX encoding deteriorates
Solution Approach 1:
The system merges multiple VAD decisions using logical AND or OR operations to achieve high sensitivity in the final output. The logical combination allows each individual VAD to use moderate thresholds while the combined result achieves high speech detection coverage, preventing unnecessary DTX activation and maintaining encoding efficiency
Data Source
AI summary
A signal activity detector (SAD) combines at least three decision signals to generate a combined decision signal as input to a hangover addition circuit of the SAD. Each of the decision signals indicating whether or not activity is detected in the input signal according to respective decision criteria. The SAD sends the combined decision signal to the hangover addition circuit to generate a final decision signal of the SAD as to whether or not activity is detected in the input signal.