Voice Activity Detection Using Temporal Audio Feature Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice activity detection methods struggle to accurately distinguish speech signals from non-speech signals in noisy environments due to the assumption of noise stationarity, leading to poor accuracy in distinguishing time and frequency domain features.
Innovation Solution
A neural network model is employed to perform high-level feature extraction and temporal modeling on audio features, using a combination of convolutional and residual network layers to enhance feature robustness and distinguishability, thereby improving voice activity detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If traditional voice activity detection methods assume noise stationarity to simplify processing, then device complexity is reduced, but measurement precision of speech signal distinction deteriorates in noisy environments
Solution Approach 1:
The patent transforms the audio signal from time-domain to frequency-domain representation using Fourier transform, changing the parameter domain to enable better noise discrimination. This parameter transformation allows the system to identify and remove noisy frequency components while preserving speech signals, resolving the contradiction between processing simplicity and detection accuracy in noisy environments
Solution Approach 2:
The patent introduces an intermediary processing stage that includes: (1) computing power spectral density from the Fourier transformed signal, (2) comparing against a noise threshold to identify noisy components, and (3) selectively removing or attenuating identified noisy frequency components. This intermediary process acts as a mediator between raw audio input and final voice activity detection, improving measurement precision without significantly increasing overall system complexity
2Device complexity
If traditional methods use simple time and frequency domain feature extraction, then device complexity is reduced, but manufacturing precision of feature distinguishability deteriorates
Solution Approach 1:
The patent segments the frequency spectrum into distinct components by applying Fourier transform and analyzing power spectral density. This segmentation separates speech signal frequencies from noise frequencies in the frequency domain, enabling precise feature distinguishability. The segmented frequency components can then be independently evaluated for voice activity detection, improving manufacturing precision of feature distinguishability
Solution Approach 2:
The patent transitions from one-dimensional time-domain analysis to two-dimensional time-frequency domain analysis by applying Fourier transform. This dimensionality change adds the frequency dimension, allowing the system to distinguish speech from noise based on their different spectral characteristics. The additional dimensional information significantly improves feature distinguishability without proportionally increasing device complexity
Data Source
AI summary
A voice activity detection method includes; obtaining a target audio feature of a target audio signal, inputting the target audio feature into a first network layer of a target model to obtain a first feature map including N first channels, inputting the first feature map into a second network layer of the target model to obtain a second feature map including N second channels, and outputting a voice activity detection category based on the second feature map. Each first channel includes one target feature matrix, and each target feature matrix is obtained by the first network layer by performing high-level feature extraction on the target audio feature. Each second channel corresponds to one first channel, each second channel includes one target feature value, and each target feature value is obtained by the second network layer by performing temporal modeling on a corresponding target feature matrix.