Noise Suppression Circuit Using Segmented Voice Activity Detectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing noise suppression technologies struggle to effectively distinguish and remove non-stationary noise, such as babble noise, from audio signals, especially when it overlaps with intended speech, leading to poor audio quality in noisy environments.
Innovation Solution
A noise suppression circuit that employs multiple voice activity detectors, including a Gaussian Mixture Model voice activity detector, to differentiate between stationary and non-stationary noise types, using a combination of noise attenuation and subtraction circuits to selectively address each type of noise, while minimizing audible artifacts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple voice activity detectors and noise reduction circuits are used to distinguish and remove different noise types, then noise suppression effectiveness is improved, but device complexity increases
Solution Approach 1:
The noise suppression circuit is segmented into multiple specialized components: a spectrum domain voice activity detector for stationary noise, a time domain voice activity detector for non-stationary noise, a Gaussian Mixture Model voice activity detector for babble noise, and corresponding specialized noise reduction circuits for each noise type. This segmentation allows each component to focus on specific noise characteristics, improving overall effectiveness while managing complexity through functional specialization.
Solution Approach 2:
The system changes detection parameters by employing multiple voice activity detectors that analyze the audio signal from different domains (spectrum domain, time domain) and using different detection approaches (Gaussian Mixture Model). Each detector uses different parameter sets to identify specific noise types, enabling the system to adapt to varying noise conditions without requiring a single complex universal detector.
2Reliability
If noise reduction techniques are applied during periods when speech is not present, then stationary noise is reduced, but non-stationary noise overlapping with speech cannot be effectively removed
Solution Approach 1:
The noise suppression circuit is designed with multi-functionality to handle both stationary and non-stationary noise types. The system incorporates multiple voice activity detectors that can identify different noise types (stationary, non-stationary, babble noise) and corresponding noise reduction circuits that can process each type appropriately. This universal design allows the system to adapt to varying noise conditions and effectively remove noise regardless of whether it overlaps with speech or occurs during silent periods.
Solution Approach 2:
The system dynamically adapts its noise reduction approach based on the detected noise type. When stationary noise is detected, the spectrum domain detector and associated circuit are activated. When non-stationary noise is detected, the time domain detector and its circuit are engaged. When babble noise is detected, the Gaussian Mixture Model detector and corresponding circuit are activated. This dynamic adaptation allows the system to optimize performance for each specific noise condition rather than using a static single-approach system.
3Speed
If static models are used for instantaneous acoustic analysis, then processing speed is improved, but effectiveness against non-stationary noise deteriorates
Solution Approach 1:
The detection system is segmented into multiple specialized detectors: a spectrum domain voice activity detector for stationary noise analysis, a time domain voice activity detector for non-stationary noise analysis, and a Gaussian Mixture Model voice activity detector for babble noise. Each detector is optimized for specific noise types and can process signals at appropriate speeds for their intended purpose, avoiding the need for a single slow comprehensive model.
Solution Approach 2:
The system changes the detection parameters and analysis methods based on the noise type being detected. For stationary noise, spectral parameters are analyzed. For non-stationary noise, time-domain parameters are emphasized. For babble noise, statistical parameters from the Gaussian Mixture Model are used. This parameter adaptation allows each detector to operate at optimal processing speed for its specific function while maintaining high effectiveness for the targeted noise type.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
A noise suppression circuit for use in an audio signal processing circuit is provided. The noise suppression circuit includes a plurality of different types of noise activity detectors, which are each adapted for detecting the presence of a different type of noise in a received signal. The noise suppression circuit further includes a plurality of different types of noise reduction circuits, which are each adapted for removing a different type of detected noise, where each noise reduction circuit respectively corresponds to one of the plurality of noise activity detectors. The respective noise reduction circuit is then selectively activated to condition the received signal to reduce the amount of the detected types of noise, when each one of the plurality of noise activity detectors detects the presence of a corresponding type of noise in the received signal.