Sound Signal Processing Using Group Convolution for Noise Suppression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Noise interference in sound signals collected by terminal devices significantly reduces the clarity and intelligibility of speeches in communication applications and the recognition rate of speech recognition systems in intelligent human-machine interaction systems, affecting user experience.
Innovation Solution
A sound signal processing method using a pre-trained sound processing model with at least one preset convolution layer, performing convolution operations on sound spectrum feature maps with specific kernel groups to generate pure audio data, reducing calculation resources and improving processing speed while maintaining accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional noise suppression methods are used, then noise can be reduced to some extent, but the processing speed is slow and calculation resources are excessive
Solution Approach 1:
The patent changes the mathematical parameters of the convolution operation by using group convolution with specific kernel groupings. The first convolution kernel group processes the first sound spectrum feature map to obtain a second sound spectrum feature map, and the second convolution kernel group combines feature maps to obtain a third sound spectrum feature map. This parameter change enables faster computation while maintaining noise suppression effectiveness.
Solution Approach 2:
The patent segments the convolution operation into multiple kernel groups that process different parts of the sound spectrum feature map independently. The first convolution kernel group handles one aspect of the feature map, while the second convolution kernel group handles another aspect, allowing parallel processing and reducing overall computation time while maintaining effective noise suppression.
2Measurement precision
If complex processing algorithms are used to improve noise suppression, then noise reduction accuracy improves, but calculation resources increase
Solution Approach 1:
The patent optimizes the convolution operation parameters by organizing kernels into specific groups. The first convolution kernel group and second convolution kernel group are configured to process feature maps with specific dimensional relationships, achieving accurate noise reduction while minimizing unnecessary computational operations and reducing energy consumption.
Solution Approach 2:
The patent applies convolution operations only to the necessary parts of the sound spectrum feature map through strategic kernel grouping. Rather than processing the entire feature map uniformly, the first and second convolution kernel groups target specific regions and relationships, achieving sufficient noise reduction accuracy with reduced computational effort.
3Ease of operation
If real-time processing is implemented, then user experience improves, but processing speed requirements increase
Solution Approach 1:
The patent segments the audio processing into independent kernel group operations that can execute in parallel. The first convolution kernel group and second convolution kernel group operate on different feature map relationships simultaneously, enabling real-time processing of sound signals while maintaining high processing speed requirements for user experience.
Solution Approach 2:
The patent modifies the convolution operation parameters to optimize for real-time processing. By configuring the first and second convolution kernel groups with specific dimensional relationships and processing patterns, the system achieves real-time noise suppression without compromising processing speed, thereby improving user experience.
Data Source
AI summary
A sound signal processing method, an electronic device, and computer-readable medium are provided. The method includes: importing first frequency spectrum data corresponding to first audio data into a pre-trained sound processing model to obtain a processing result; and generating, based on the processing result, pure audio data corresponding to the first audio data. The sound processing model includes at least one preset convolution layer, and operations performed by using the preset convolution layer includes: performing, based on a first convolution kernel group, a convolution operation on a first sound spectrum feature map inputted into the preset convolution layer, to obtain a second sound spectrum feature map; and combining, based on a second convolution kernel group, the obtained second sound spectrum feature map, to obtain a third sound spectrum feature map corresponding to the second convolution kernel group.