Voiced Sound Interval Detection Using Multidimensional Vector Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voiced sound interval detection technologies face challenges in accurately classifying voiced sound intervals when sound source volume varies or when the number of sound sources is unknown, especially when different microphones are used together, leading to poor voice interval detection performance.
Innovation Solution
A voiced sound interval detection device and method that calculates a multidimensional vector series from power spectrum time series, clusters the vectors, and determines voiced sound intervals by calculating a signal-to-noise ratio using center vectors of noise and voice clusters, allowing for accurate detection regardless of sound source volume variations or unknown sound sources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional S/N ratio calculation methods are used for voiced sound interval detection, then the detection process is simple, but the detection precision deteriorates when sound source volume varies or when different microphones are used together
Solution Approach 1:
The patent transforms the traditional scalar S/N ratio calculation into an M-dimensional vector space operation. By representing power spectrum data from multiple microphones as vectors in M-dimensional space and performing clustering in this expanded dimensionality, the system achieves more accurate voiced sound interval detection that accounts for variations in sound source volume and microphone characteristics, while maintaining computational efficiency through vectorized operations.
2Measurement precision
If clustering is performed in M-dimensional space to account for microphone differences, then detection accuracy improves, but computational complexity increases
Solution Approach 1:
The patent segments the power spectrum data from multiple microphones into separate vector components, where each dimension corresponds to one microphone. By performing clustering operations on these segmented vectors in M-dimensional space, the system accurately captures microphone-specific characteristics and sound source variations. The segmentation enables independent analysis of each microphone's contribution while maintaining the computational efficiency of vector operations.
3Adaptability or versatility
If the system handles unknown number of sound sources, then adaptability improves, but the difficulty of detecting and measuring increases
Solution Approach 1:
The patent changes the parameter representation from traditional scalar S/N ratios to M-dimensional power spectrum vectors. This parameter transformation enables the system to naturally handle an unknown number of sound sources by capturing the full spectral characteristics across all microphones. The vector-based approach allows the clustering algorithm to automatically adapt to varying numbers of sound sources without requiring prior knowledge, as the M-dimensional space inherently represents all possible sound source configurations.
Data Source
AI summary
This invention provides a voiced sound interval detection device which enables appropriate detection of a voiced sound interval of an observation signal even when a volume of sound from a sound source varies or when the number of sound sources is unknown or when different kinds of microphones are used together.