Spectral Smoothing for Noise Reduction in Audio Signals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional electronic devices face challenges in accurately reducing noise in audio data during communication sessions, leading to signal quality degradation and distortion due to incorrect noise frame estimations and the introduction of artifacts like musical noise and reverberation.
Innovation Solution
The system performs noise reduction using curve fitting and Savitzky-Golay filtering to smooth gain functions, generating mask data that isolates speech by frame-by-frame processing of single-channel noisy acoustic signals, improving signal-to-noise ratio and reducing distortion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If conventional noise reduction methods are used, then noise suppression is achieved, but signal quality degradation and distortion occur due to incorrect noise frame estimations
Solution Approach 1:
The patent changes the parameter estimation approach by using spectral smoothing techniques to estimate noise power spectrum more accurately. Instead of relying on conventional VAD-based frame classification which causes distortion, the invention continuously estimates noise parameters across frequency bins using smoothed spectral data, thereby maintaining signal quality while achieving noise suppression.
Solution Approach 2:
The patent introduces spectral smoothing as an intermediary process between raw audio input and noise reduction output. By applying smoothing windows and averaging techniques to the spectral data before noise estimation, the system creates a more reliable intermediate representation that reduces the harmful effects of incorrect frame classifications and improves overall signal quality.
2Object-affected harmful factors
If conventional noise reduction algorithms are applied, then noise is reduced, but external artifacts like musical noise and reverberation are introduced
Solution Approach 1:
The patent applies partial smoothing across frequency bins rather than uniform processing. By selectively smoothing certain frequency regions and using different smoothing factors for different bins, the system achieves noise reduction in problematic frequency ranges while preserving natural characteristics in others, thereby reducing artifact generation.
Solution Approach 2:
The patent implements dynamic noise estimation that adapts to changing acoustic environments. The smoothing parameters and noise estimation are continuously updated based on current spectral conditions rather than using fixed conventional algorithms, allowing the system to maintain low artifact levels across varying noise conditions and speech activities.
3Speed
If frame-by-frame processing is used, then real-time processing is achieved, but distortion occurs due to sudden gain changes between frames
Solution Approach 1:
The patent performs preliminary spectral smoothing and noise estimation before applying gain adjustments in each frame. By pre-smoothing the spectral data and estimating noise parameters in advance, the system prepares gain values that are inherently more stable and less prone to sudden changes, thereby reducing distortion while maintaining real-time processing capability.
Solution Approach 2:
The patent applies periodic smoothing operations across frames using overlapping windows and continuous spectral updates. This periodic application of smoothing with appropriate windowing functions ensures that gain changes between frames are gradual and continuous, eliminating the sudden transitions that cause distortion while preserving real-time processing speed.
Data Source
AI summary
A system configured to perform low input-output latency noise reduction in a frequency domain is provided. The real-time noise reduction algorithm performs frame by frame processing of a single-channel noisy acoustic signal to estimate a gain function. Accurate noise power estimates are achieved with the help of minimum statistics approach followed by a voice activity detector. The noise power and gain values are smoothed to remove any external artifacts and avoid background noise modulations. The gain values for individual frequency bands are weighted and smoothed to reduce distortion. To obtain distortionless output speech, the system performs curve fitting by separating the frequency bands into multiple groups and applying a Savitzky-Golay filter to each group. The final gain values generated by these filters are multiplied with the noisy speech signal to obtain a clean speech signal.