Audio Signal Processing with DNN Feature Restoration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech processing systems face challenges in reducing audible ambient noise and improving speech intelligibility, particularly in environments with significant distance between speakers and microphones, where linear noise reduction algorithms are insufficient and require precise tuning to achieve desired signal-to-noise ratios without speech distortion.
Innovation Solution
A system employing a deep neural network (DNN)-based feature restoration algorithm, comprising STFT, convolutional layers, LSTM layers, dense layers, and transposed convolutional layers, is used to process audio signals, transforming them between amplitude and frequency domains to generate an output signal that reduces noise and enhances speech clarity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If non-linear post filtering algorithm is used to enhance noise reduction levels, then signal-to-noise ratio is improved, but speech distortion increases
Solution Approach 1:
A feature restoration algorithm acts as an intermediary between the non-linear post filtering algorithm and the final output. The post filter reduces noise but introduces distortion, while the feature restoration algorithm (using deep neural networks with STFT, convolutional layers, LSTM layers, and transposed convolutional layers) restores speech features and removes distortion artifacts, thereby maintaining both high SNR and low distortion
Solution Approach 2:
The patent replaces traditional mechanical tuning and simple filtering mechanisms with a deep learning-based feature restoration system. Instead of manually tuning filters to balance noise reduction and distortion, the system uses trained neural networks (including LSTM for temporal dependencies and transposed convolutional layers for feature reconstruction) that automatically learn and restore speech features, eliminating the need for precise manual tuning
2Reliability
If spectral subtraction algorithms are employed in post filtering module, then noise reduction is enhanced, but speech features are removed causing distortion
Solution Approach 1:
The system deliberately discards potentially distorted speech features during post filtering (allowing aggressive noise reduction), then recovers and restores these features using the feature restoration algorithm. The deep neural network learns to identify and reconstruct speech features that were removed or distorted by spectral subtraction, thereby recovering lost information while maintaining noise reduction benefits
3Reliability
If precise tuning is applied to reach target SNR with minimal speech distortion, then speech quality is improved, but system complexity and tuning difficulty increase
Solution Approach 1:
The feature restoration algorithm using deep neural networks performs self-service by automatically adapting to different acoustic environments and speech characteristics without manual intervention. The LSTM layers learn temporal patterns and the transposed convolutional layers reconstruct features autonomously, allowing the system to maintain optimal speech quality across varying conditions without requiring precise manual tuning for each scenario
Data Source
AI summary
A system and method for processing an audio input signal includes a microphone, a controller, and a communication link that may be coupled to a remote speaker. The microphone captures the audio input signal and communicates the audio input signal to the controller, and the controller is coupled to the communication link. The controller includes executable code to generate, via a linear noise reduction filtering algorithm, a first resultant based upon the audio input signal, and generate, via non-linear post filtering algorithm, a second resultant based upon the first resultant. An audio output signal is generated based upon the second resultant employing a feature restoration algorithm. The audio output signal is communicated, via the communication link, to a speaker that may be at a remote location.