Frequency-Domain Audio Noise Reduction for Embedded Real-Time Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep learning-based audio noise reduction models are too large and complex, making them unsuitable for deployment on embedded terminals and failing to meet real-time requirements, while traditional methods struggle to effectively suppress non-stationary noise, causing audio distortion.
Innovation Solution
An audio noise reduction method using a depth neural network with an encoder, recurrent neural network layer, and decoder module, performing amplitude and phase feature extraction and reconstruction, and a noise-reduced signal mask generation, optimized through a loss function for efficient deployment on embedded terminals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional digital signal processing methods are used for audio noise reduction, then the processing speed is fast and the model size is small, but they cannot effectively suppress non-stationary noise and cause audio distortion
Solution Approach 1:
The patent segments the audio signal processing into distinct amplitude and phase components, processing them separately through dedicated neural network branches. This segmentation allows the model to handle non-stationary noise more effectively while maintaining computational efficiency suitable for embedded deployment.
Solution Approach 2:
The patent transforms the audio signal from time domain to frequency domain representation, adding a frequency dimension to the processing. This dimensional transformation enables the model to capture spectral characteristics of non-stationary noise, improving noise reduction effectiveness while using a compact architecture.
2Reliability
If existing deep neural network models are used for audio noise reduction, then the noise reduction effectiveness is improved, but the model size becomes excessively large and cannot be deployed on embedded terminals
Solution Approach 1:
The model is segmented into separate amplitude and phase processing branches, each with dedicated feature extraction networks. This segmentation reduces the overall model size by avoiding redundant computations while maintaining effective noise reduction performance for non-stationary noise.
Solution Approach 2:
The patent extracts and processes only the essential amplitude and phase features needed for noise reduction, discarding redundant information. This extraction approach reduces model size while preserving the key characteristics necessary for effective non-stationary noise suppression.
3Reliability
If existing deep neural network models are used for audio noise reduction, then the noise reduction effectiveness is improved, but the processing time increases and real-time requirements are not met
Solution Approach 1:
By segmenting the processing into parallel amplitude and phase branches with dedicated feature extractors, the model achieves efficient computation. Each branch processes its specific features independently, reducing overall processing time while maintaining effective non-stationary noise suppression.
Solution Approach 2:
The patent applies partial action by focusing computational resources only on the essential amplitude and phase features required for noise reduction, rather than processing all possible audio characteristics. This selective approach reduces processing time while maintaining effective noise suppression performance.
Data Source
AI summary
This application provides an audio noise reduction method, an audio noise reduction model training method, and an electronic device. In the audio noise reduction method, a frequency domain audio is obtained by converting an audio to be noise-reduced from time domain to frequency domain frame by frame. The concatenated features of the frequency domain audio are generated by performing an amplitude concatenation for each frame of the frequency domain audio and a phase concatenation for each frame. The concatenated features are putted into an audio noise reduction model to obtain a signal mask, and then the signal mask is processed to obtain a noise-reduced audio. By implementing the audio noise reduction method, the effect of audio noise reduction is improved, and the structure of the noise reduction model is simplified.