Multi-layer ML Noise Filtering for Real-time Audio
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing active noise cancellation technologies are ineffective in public spaces, outdoor areas, and near highways or airports, and high-accuracy machine learning models for noise filtering are computationally expensive, making them difficult to implement in real-time on lightweight devices used for audio communications.
Innovation Solution
A multi-layered machine learning approach using computationally inexpensive preliminary classifiers to filter out low-probability background noise events, followed by a more accurate model for further classification and attenuation, allowing for efficient noise reduction in real-time communications without disrupting audio quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-accuracy machine learning models are used for noise filtering, then noise classification accuracy is improved, but computational complexity increases making real-time implementation difficult
Solution Approach 1:
The patent segments the machine learning noise filtering process into multiple layers: a first layer using a computationally inexpensive preliminary classifier to filter out obvious background noise events, and a second layer using a more accurate but computationally intensive model only for events that pass the first filter. This segmentation reduces overall computational complexity while maintaining high accuracy for relevant noise events.
Solution Approach 2:
The patent applies partial action by using a simplified classification approach for the majority of audio events and reserving the more computationally demanding accurate classification only for a subset of events that require it. This selective application of computational resources maintains high accuracy where needed while reducing overall computational burden.
2Productivity
If computationally inexpensive preliminary classifiers are used, then processing speed is improved, but noise detection accuracy decreases leading to false positives
Solution Approach 1:
The patent introduces an intermediary multi-layered classification system where the preliminary classifier acts as a first-stage filter that quickly identifies and eliminates obvious background noise events. Events that are not clearly identified as background noise by the first classifier are passed to a second, more accurate classifier for further evaluation, thus reducing false positives while maintaining processing speed.
Solution Approach 2:
The patent replaces a single mechanical classification decision with a multi-stage classification process, substituting the need for a single complex accurate classifier with a sequence of simpler classifiers that collectively achieve both speed and accuracy.
3Ease of manufacture
If legacy active noise cancellation technologies are used, then implementation is simple, but applicability is limited to small enclosed spaces and continuous low frequency sounds
Solution Approach 1:
The patent changes the fundamental parameters of noise cancellation by transitioning from frequency-based signal processing methods (legacy ANC) to machine learning-based event classification. This parameter change enables the system to handle a broader range of noise types including intermittent and high-frequency sounds, while maintaining implementation feasibility through the use of trained classification models.
Data Source
AI summary
This disclosure relates to solutions for eliminating undesired audio artifacts, such as background noises, on an audio channel. A process for implementing the technology can include receiving a set of audio segments, analyzing the segments using a first ML model to identify a first probability of unwanted background noises in the segments, and if the first probability exceeds a threshold, analyzing the segments using a second ML model to determine a second probability that the one or more background features exist in the segments. In some aspects, the process can include attenuating audio artifacts in the segments, if the second probability exceeds a second threshold. In some implementations, dynamic time stretching and shrinking can be applied to the noise attenuation. Systems and machine-readable media are also provided.