Voice Activity Detection Using Multi-Model Probability Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice activity detection (VAD) technologies lack accuracy in determining the presence or absence of human speech without pre-selecting the appropriate machine learning model for balancing parameters such as volume or wavelength.
Innovation Solution
An information processing device and method that segments audio data into time zones, applies multiple machine learning models (e.g., HMM, SVM) to generate probability time-series data, and uses a combined model to determine the presence probability of human speech, improving accuracy without pre-determining the model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple machine learning models are used to improve detection accuracy, then measurement precision improves, but device complexity increases
Solution Approach 1:
The patent implements dynamic model selection where the system automatically chooses the most appropriate machine learning model based on real-time analysis of audio characteristics such as signal-to-noise ratio, spectral features, and temporal patterns. This dynamic adaptation allows the system to maintain high detection accuracy across varying acoustic environments without requiring manual configuration or complex fixed architectures.
Solution Approach 2:
The system changes operational parameters by selecting different machine learning models according to detected acoustic conditions. When noise levels are high, one model configuration is selected; when speech characteristics differ, another model is chosen. This parameter-based model selection resolves the contradiction by adapting the system's computational approach to match the input characteristics, maintaining precision while managing complexity through conditional logic rather than permanent multi-model architecture.
2Reliability
If multiple machine learning models are applied to handle varying noise levels, then reliability improves, but device complexity increases
Solution Approach 1:
The patent employs dynamic model selection that adapts to varying noise levels by analyzing acoustic environment characteristics in real-time. The system monitors signal properties and automatically switches between pre-trained machine learning models optimized for different noise conditions, ensuring reliable detection across diverse environments without requiring a permanently active complex multi-model system.
Solution Approach 2:
Multiple machine learning models are pre-trained offline for different acoustic scenarios and noise levels. During runtime, the system selects from these pre-prepared models based on current conditions, rather than dynamically training or switching between complex architectures in real-time. This preliminary preparation maintains reliability across varying conditions while keeping the runtime processing complexity manageable.
3Manufacturing precision
If machine learning models are trained for specific parameters, then manufacturing precision improves, but adaptability decreases
Solution Approach 1:
The patent creates a universal framework where multiple specialized machine learning models are integrated into a single system. Each model is trained with high precision for specific acoustic conditions or parameter ranges, but the system as a whole is versatile because it can select the appropriate specialized model based on the input characteristics. This multi-functional architecture resolves the contradiction by combining specialized precision with general adaptability through model selection rather than requiring a single general-purpose model.
Solution Approach 2:
The patent segments the detection task by dividing it into multiple specialized machine learning models, each optimized for specific acoustic conditions, noise levels, or speech characteristics. Instead of one general model attempting to handle all scenarios, the system segments the problem space and assigns specialized models to different segments, then selects the appropriate segment's model based on current conditions. This segmentation maintains high precision for each specialized model while providing overall adaptability through the segmented architecture.
Data Source
AI summary
In order to improve accuracy for detecting presence or absence of a target object. A time-series segmentation unit 102 creates first time-series data by segmenting processing target data into each frame of “n” time zones. Each of first determination units 103 creates “m” second time-series data by determining each frame of the first time-series data using “m” models having different characteristics. A second determination unit 104 creates a second determination result as a presence probability of the target object for a set of second time-series data including n×m data.