Adaptive Noise Suppression Using Dual ML Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing noise suppression technologies in video conferencing often result in sound distortion due to inaccurate noise identification, either failing to suppress noise or removing essential sounds, leading to suboptimal audio quality in diverse environments.
Innovation Solution
The implementation of a two-stage machine learning approach using convolutional neural networks to classify environments and detect scenarios, allowing for adaptive noise suppression based on the spatial relationship between noise sources and users, enhancing audio quality by selectively suppressing background noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If existing noise suppression technologies are used, then noise can be suppressed, but sound distortion occurs due to inaccurate noise identification
Solution Approach 1:
The patent segments the audio processing task into two distinct stages: first classifying the environment type (office, non-office, unknown), then detecting specific scenarios within that environment. This segmentation allows each stage to focus on specific features, improving overall accuracy in identifying noise sources while preserving important sounds.
Solution Approach 2:
The patent adds a spatial dimension to noise suppression by detecting the location of noise sources relative to the user and adjusting suppression levels accordingly. Sounds from distant locations are suppressed more than sounds from nearby locations, creating a spatially-aware noise suppression system that reduces distortion while maintaining audio quality.
2Object-affected harmful factors
If noise suppression is applied, then background noise is reduced, but essential sounds are removed leading to suboptimal audio quality
Solution Approach 1:
The patent applies different noise suppression levels to different sound sources based on their spatial location and importance. Instead of uniform suppression, the system selectively applies suppression to distant noise sources while preserving nearby important sounds, ensuring local optimization of audio quality in different spatial zones.
Solution Approach 2:
The system continuously monitors the audio environment, classifies environments and scenarios, and adjusts noise suppression levels in real-time based on detected conditions. This feedback loop ensures that suppression levels are dynamically optimized to maintain audio quality while reducing background noise.
3Measurement precision
If a two-stage machine learning approach is implemented, then noise identification accuracy is improved, but device complexity increases
Solution Approach 1:
The complex task of noise identification is segmented into two manageable stages: environment classification and scenario detection. Each stage uses specialized machine learning models trained for specific purposes, making the overall system more efficient and accurate while keeping individual components relatively simple.
Solution Approach 2:
The system performs preliminary environment classification before conducting detailed scenario detection. This preliminary action narrows down the search space and allows the second stage to focus on specific scenarios relevant to the detected environment, improving efficiency and reducing computational complexity.
Data Source
AI summary
Examples of noise suppression controls are described herein. In some examples, an electronic device includes a processor to classify, using a first machine learning model, an environment based on video of the environment to produce a classification. In some examples, the processor is to detect, using a second machine learning model, a situation in the environment based on the video to produce a detection. In some examples, the processor is to control noise suppression on audio captured from the environment based on the classification and the detection.