Neural Network Background Sound Elimination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for eliminating background sound in noisy environments, such as acoustic echo cancellation, require repeated learning and are complex, costly, and provide poor user experience.
Innovation Solution
A method using neural network training based on initial audio data sets and background sound fusion processing to generate a neural network model for eliminating background sound, which can be applied universally without additional training, improving communication quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If acoustic echo cancellation is used to eliminate background sound, then background sound elimination is achieved, but the method becomes complex and requires repeated learning
Solution Approach 1:
The patent replaces traditional acoustic echo cancellation methods with a neural network-based approach. The neural network model learns to eliminate background sound through training on audio data, substituting the mechanical signal processing methods with an intelligent system that automatically adapts to different acoustic environments without requiring complex repeated learning.
Solution Approach 2:
The patent performs preliminary training of the neural network model using background sound fusion processing on audio data sets before actual use. This preliminary action creates a pre-trained model that can directly eliminate background sound in various scenarios without requiring repeated learning during operation, thereby reducing complexity while maintaining reliability.
2Reliability
If traditional background sound elimination methods are used, then background noise is removed, but the cost increases and user experience deteriorates
Solution Approach 1:
The patent substitutes expensive traditional background noise elimination systems with a neural network model that can be trained and deployed more efficiently. The neural network approach reduces implementation costs by leveraging automated learning processes rather than requiring complex hardware or repeated manual tuning, while maintaining effective background noise removal.
3Adaptability or versatility
If separate model training is performed for each scenario, then adaptation to specific environments is achieved, but the process becomes time-consuming and costly
Solution Approach 1:
The patent creates a universal neural network model through background sound fusion processing that can adapt to multiple different scenarios simultaneously. Instead of training separate models for each environment, the fusion processing enables a single model to handle various acoustic conditions, thereby maintaining environmental adaptability while eliminating the need for repeated time-consuming training processes.
Solution Approach 2:
The patent performs comprehensive background sound fusion processing in advance to create a pre-trained neural network model that is already adapted to multiple environments. This preliminary action eliminates the need for scenario-specific training during deployment, significantly reducing training time while preserving the model's ability to adapt to different acoustic scenarios.
Data Source
AI summary
The present disclosure provides a method and a device for eliminating background sound, and a terminal device. The method includes: obtaining an initial audio data set; performing background sound fusion processing on the initial audio data set to obtain training sample data; performing neural network training based on the training sample data and the initial audio data set to generate an initial neural network model for eliminating background sound; and performing background sound elimination on audio data to be processed based on the initial neural network model for eliminating background sound.