Voice Activity Detection Using Dynamic Feature Weighting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice activity detection (VAD) systems face challenges in accurately distinguishing speech from noise, especially in complex acoustic environments with dynamic signal energy levels and high background noise, leading to errors in speech recognition and quality degradation.
Innovation Solution
A method that combines energy-based and model-based feature vectors using a dynamic weighting factor to adapt to environmental changes, where energy-based feature vectors from different frequency bands and acoustic models are used in conjunction with preclassifiers and classifiers to classify speech and noise events effectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If threshold comparison techniques are used for voice activity detection, then the system is simple to implement, but it fails to handle complex acoustic situations with dynamic signal energy levels and background sounds
Solution Approach 1:
The patent combines multiple feature extraction approaches (energy-based features, zero-crossing rate, autocorrelation coefficients, and model-based features like MFCC) into a unified VAD system. This integration allows the system to leverage the simplicity of threshold comparison while incorporating the robustness of model-based techniques, thereby improving detection accuracy in complex acoustic environments without sacrificing implementation feasibility.
Solution Approach 2:
The system uses a composite feature vector that integrates multiple types of acoustic features (energy, spectral, temporal) into a single comprehensive representation. This composite approach enables the VAD system to capture both simple energy variations and complex spectral patterns, resolving the contradiction between implementation simplicity and detection reliability.
2Reliability
If model based VAD techniques are used, then the system works better in noisy conditions, but their dependency on a single language reduces their functionality
Solution Approach 1:
The patent employs universal acoustic features (energy, zero-crossing rate, autocorrelation, MFCC) that are language-independent and can capture general speech characteristics across different languages. The system processes speech and non-speech events based on acoustic patterns rather than language-specific phonemes, enabling multilingual applicability while maintaining robustness in noisy conditions.
Solution Approach 2:
The system segments the feature extraction process into multiple independent components (energy-based features, spectral features, temporal features) that can be processed separately and combined. This segmentation allows each feature type to contribute independently to the detection decision, enabling the system to maintain language independence while achieving reliable performance in noisy environments through the combined power of multiple feature types.
3Reliability
If adaptive thresholds are used to improve VAD performance, then the system can handle dynamic signal energy levels, but it increases system complexity
Solution Approach 1:
The patent implements adaptive thresholding mechanisms that dynamically adjust detection thresholds based on the statistical properties of the input signal. The system computes energy thresholds and decision thresholds adaptively from the training data and updates them during operation to match changing acoustic conditions. This dynamic adaptation enables the system to handle varying signal energy levels and background noise without requiring manual threshold tuning for each environment.
Solution Approach 2:
The VAD system performs self-adaptation by automatically adjusting its thresholds and parameters based on the statistical characteristics of the input signal. The training phase automatically determines optimal thresholds from training data, and the recognition phase continuously adapts to changing conditions without external intervention. This self-service capability reduces the need for manual configuration and maintains reliability across diverse acoustic environments.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Discrimination between at least two classes of events in an input signal is carried out in the following way. A set of frames containing an input signal is received, and at least two different feature vectors are determined for each of said frames. Said at least two different feature vectors are classified using respective sets of preclassifiers trained for said at least two classes of events. Values for at least one weighting factor are determined based on outputs of said preclassifiers for each of said frames. A combined feature vector is calculated for each of said frames by applying said at least one weighting factor to said at least two different feature vectors. Said combined feature vector is classified using a set of classifiers trained for said at least two classes of events.