Voice Activity Detection Using Dominant Gaussian Distribution Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional voice activity detection methods process large amounts of information by using all Gaussian distributions, which increases processing time, especially when dealing with multi-channel acoustic signals.
Innovation Solution
A voice activity detection apparatus that generates non-speech and speech probabilistic models adapted to a noise environment by using silence and clean-speech GMMs, calculating probabilities of dominant distributions only, and employs a noise suppressor to filter out noise in the acoustic signal.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all Gaussian distributions are used in voice activity detection, then the accuracy of speech period estimation is improved, but the processing time increases
Solution Approach 1:
The patent extracts and processes only the dominant Gaussian distributions from the complete GMM set. By identifying and selecting only the most significant distributions (those with highest probabilities) for processing, the system maintains speech detection accuracy while reducing the computational burden of processing all distributions.
Solution Approach 2:
Instead of processing all Gaussian distributions equally, the patent applies partial action by processing only a subset of the most important distributions. This selective approach processes fewer elements (partial) while focusing computational resources on the most relevant distributions, thereby reducing overall processing time without significantly compromising detection accuracy.
2Measurement precision
If individual processing is performed in each channel for multi-channel acoustic signals, then the detection accuracy is improved, but the amount of information to be processed increases
Solution Approach 1:
The patent segments the processing task by channel, applying individual processing to each channel's dominant distributions. This segmentation allows the system to handle multi-channel signals with the same efficiency as single-channel signals, as each channel is processed independently with reduced computational requirements.
3Reliability
If all Gaussian distributions are processed in multi-channel signals, then comprehensive speech detection is achieved, but the processing speed decreases
Solution Approach 1:
The patent extracts only the dominant Gaussian distributions from each channel's GMM before processing. By removing unnecessary distributions and keeping only the most significant ones, the system maintains reliable speech detection across all channels while dramatically reducing the total amount of data that requires processing.
Solution Approach 2:
The patent applies partial processing by handling only a subset of distributions (the dominant ones) in each channel rather than processing all distributions. This approach achieves sufficient detection reliability through selective processing of the most informative distributions, thereby improving overall processing speed.
Data Source
AI summary
The processing efficiency and estimation accuracy of a voice activity detection apparatus are improved. An acoustic signal analyzer receives a digital acoustic signal containing a speech signal and a noise signal, generates a non-speech GMM and a speech GMM adapted to a noise environment, by using a silence GMM and a clean-speech GMM in each frame of the digital acoustic signal, and calculates the output probabilities of dominant Gaussian distributions of the GMMs. A speech state probability to non-speech state probability ratio calculator calculates a speech state probability to non-speech state probability ratio based on a state transition model of a speech state and a non-speech state, by using the output probabilities; and a voice activity detection unit judges, from the speech state probability to non-speech state probability ratio, whether the acoustic signal in the frame is in the speech state or in the non-speech state and outputs only the acoustic signal in the speech state.