Machine Learning Noise Suppression for Mechanical Keyboards
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional noise suppression methods, such as bandpass filters, are ineffective in reducing input device noise from mechanical keyboards and other devices during electronic conferences, leading to distracting and nuisance noise interference.
Innovation Solution
A machine learning-based noise suppression system using a trained deep neural network analyzes audio feeds to identify and filter out input device activation sounds, retaining speech signals while minimizing computational delays.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If mechanical keyboards are used to provide tactile feedback and better typing experience, then typing experience is improved, but noise generation increases
Solution Approach 1:
The patent extracts and removes the harmful noise component from the audio signal while preserving the useful speech content. The system identifies and subtracts the mechanical keyboard noise spectrum from the recorded audio, effectively separating the harmful noise from the useful signal to maintain typing experience while reducing noise interference.
Solution Approach 2:
The patent converts the harmful mechanical keyboard noise into a beneficial filtering reference. By capturing and analyzing the noise spectrum from mechanical keyboards, the system creates a spectral template that can be used to identify and remove similar noise patterns from audio signals, turning the harmful noise characteristic into a useful tool for noise cancellation.
2Device complexity
If conventional bandpass filters are used to suppress noise, then implementation is simple, but noise suppression effectiveness is insufficient
Solution Approach 1:
The patent changes the approach from fixed-frequency bandpass filters to dynamic spectral subtraction based on noise templates. Instead of using static frequency ranges, the system captures the actual noise spectrum characteristics of mechanical keyboards and uses these temporal-spectral parameters to dynamically identify and remove noise, significantly improving suppression effectiveness while maintaining reasonable system complexity.
Solution Approach 2:
The patent creates a spectral copy or template of the mechanical keyboard noise pattern and uses this template to identify and subtract noise from audio signals. By copying the noise characteristics into a reusable spectral template, the system achieves effective noise suppression without requiring complex real-time analysis, balancing simplicity and effectiveness.
3Reliability
If noise suppression processing is applied during electronic conferences, then audio quality is improved, but computational delay increases
Solution Approach 1:
The patent performs preliminary action by pre-capturing and storing noise spectral templates during calibration periods when only mechanical keyboard noise is present. These pre-computed spectral templates are stored for rapid retrieval and application during actual conference sessions, eliminating the need for complex real-time noise characterization and reducing computational delay while maintaining audio quality improvement.
Data Source
AI summary
A method includes receiving sound input features representative of sound received during an electronic conference, the sound including voice and input device activation sound, receiving an input event feature indicative of the input device activation, and processing the received sound input features and input event feature via a trained model to identify a stored spectral file to be subtracted from the received sound to suppress the input device activation sound.