Divisive Normalization for Noise-Robust Spiking Audio Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing audio signal processing systems struggle with accurately distinguishing desired signals from background noise, particularly in environments with fluctuating noise levels, leading to misclassification due to the variability in spike rates caused by background noise, which conventional normalization methods fail to adequately address.
Innovation Solution
A divisive normalization method that includes averaging input spikes over a frame period, using a low-pass filter to calculate a threshold parameter, and employing an Integrate-and-Fire counter to generate normalized spikes, thereby reducing background noise to a constant level, while preserving spike order and avoiding cross-channel distortions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional normalization methods are used to handle background noise, then the system can process audio signals, but the spike rates fluctuate due to varying noise levels causing misclassification
Solution Approach 1:
The patent introduces an intermediary normalization module between the audio front-end and the classification layer. This module computes a normalization factor based on the statistical properties of the input spikes and applies it to stabilize the spike rates. The intermediary component decouples the direct relationship between fluctuating background noise and spike rate variations, allowing reliable classification even in noisy environments.
Solution Approach 2:
The normalization process uses feedback from the input spike statistics to dynamically adjust the normalization factor. By continuously monitoring the mean and variance of input spikes and adapting the normalization parameters accordingly, the system compensates for varying background noise levels and maintains stable spike rates for accurate classification.
2Reliability
If detailed normalization processing is applied to each channel, then noise suppression improves, but computational complexity and power consumption increase
Solution Approach 1:
The normalization process is segmented into efficient computational stages: spike counting, statistical moment computation, and factor application. Each channel is processed independently but uses shared statistical computations, reducing redundant calculations. The segmentation allows the system to achieve effective noise suppression while minimizing computational overhead and power consumption.
Solution Approach 2:
The patent changes the parameter representation from raw spike waveforms to compact statistical descriptors (mean, variance, and normalized spike rates). This parameter transformation reduces the data volume requiring processing and enables efficient computation of normalization factors, thereby lowering power consumption while maintaining noise suppression effectiveness.
3Reliability
If complex normalization algorithms are used, then noise handling improves, but processing speed decreases
Solution Approach 1:
The patent replaces complex iterative normalization algorithms with a direct computational approach using closed-form solutions. Instead of using heavy mechanical processing or iterative optimization, the system computes normalization factors using straightforward statistical formulas applied to spike counts, achieving both robust noise handling and high processing speed suitable for real-time audio classification.
4Reliability
If normalization is applied to stabilize spike rates, then classification reliability improves, but the structure becomes more complex
Solution Approach 1:
The normalization module is designed with multi-functionality, handling multiple tasks within a unified structure: spike rate computation, statistical moment calculation, normalization factor derivation, and application. This universal approach avoids the need for separate dedicated components for each function, reducing overall structural complexity while maintaining classification reliability.
Data Source
AI summary
A divisive normalization method, device, audio feature extractor, and a chip are disclosed. For improving the robustness against variations of the background noise, the proposed per-channel divisive normalization method comprises: obtaining an average number of input spikes over an averaging window by a low-pass filter to produce a threshold parameter, then deciding whether to enable an integrate-and fire (IAF) counter counting over a clock period of a divisive normalization module, when the count value of IAF counter reaches the threshold, resetting IAF counter and producing a single spike at the output. Compared with the prior art, the proposed method does not require any Linear Feedback Shift Register (LFSR) and is easier to implement. It also has a simpler structure, higher accuracy and better statistical performance, and a lower cost and a lower power consumption.