Platform Fan Noise Suppression for Low-Latency Microphone Audio
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing audio systems struggle to effectively reduce fan noise generated by active cooling fans, which interferes with audio signal quality, particularly in devices like laptops, despite the use of high signal-to-noise ratio microphones and software solutions, as fan noise remains discernible.
Innovation Solution
A neural network model utilizing a low-latency Short-Time Fourier Transform (LL-STFT) and a recurrent neural network (RNN) architecture, including a custom Gated Recurrent Unit (GRU) layer, to enhance microphone Signal-to-Noise Ratio (SNR) and Signal-to-Distortion-plus-Noise Ratio (SDNR) by predicting and removing platform fan noise while minimizing latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional STFT is used for noise reduction, then noise filtering capability is improved, but algorithmic latency increases to 24ms
Solution Approach 1:
The patent implements a dynamic processing pipeline that adapts the STFT parameters and neural network processing depth based on the detected fan noise characteristics. The system dynamically adjusts the frame size, hop size, and neural network model complexity to optimize the balance between noise reduction effectiveness and processing latency, achieving real-time performance with reduced 5ms latency while maintaining effective fan noise suppression.
2Measurement precision
If expensive hardware components are used to minimize fan noise impact, then audio signal quality is improved, but device cost increases
Solution Approach 1:
The patent replaces expensive hardware solutions (such as premium microphones with superior noise rejection, active noise cancellation hardware, or specialized audio processing circuits) with a software-based neural network approach. The trained neural network model processes audio signals to identify and suppress fan noise characteristics, achieving comparable or superior noise reduction performance to expensive hardware while significantly reducing device manufacturing costs.
3Object-affected harmful factors
If high pass filters are applied to remove low frequency noise, then background noise is reduced, but audio signal distortion increases
Solution Approach 1:
The patent applies localized processing to different frequency regions and time segments of the audio signal. The neural network identifies specific frequency bands contaminated by fan noise and applies targeted suppression only to those regions, while preserving the rest of the audio spectrum. This selective, localized approach reduces background noise effectively without introducing the broad-signal distortion characteristic of global high pass filtering.
4Object-affected harmful factors
If adaptive noise cancelation is used to subtract background noise, then noise reduction is improved, but computational complexity increases
Solution Approach 1:
The patent performs preliminary training of the neural network model offline using labeled datasets containing fan noise characteristics. During real-time operation, the pre-trained model executes inference with significantly reduced computational requirements compared to adaptive noise cancelation that continuously learns and adjusts parameters. The heavy computational lifting is done beforehand, enabling efficient real-time noise reduction with lower device complexity and power consumption.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
Systems and methods are provided an audio signal enhancement system that attenuates platform fan noise. Fan noise is a common type of self-noise in laptops and other devices, and fan noise can significantly degrade the quality of audio captured by built-in microphones. A neural network model is provided that enhances microphone Signal-to-Noise Ratio (SNR) and Signal-to-Distortion-plus-Noise Ratio (SDNR). The systems and methods also reduce algorithmic latency. The model architecture includes a Recurrent Neural Network, and a custom Gated Recurrent Unit layer is provided that uses fewer unique matrix weights and fewer biases and has fewer compute operations using fewer parameters. A platform self-noise suppression system is provided that eliminates low-amplitude platform self-noise signals. The model can predict when the platform fan is active, and remove the platform noise. In some examples, when the model predicts that the platform fan is not active, the model focuses on removing microphone self-noise.