Level-Dependent Noise Suppression Gain Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech enhancement algorithms struggle to effectively suppress sudden bursts of loud noise while maintaining speech quality, due to limitations in estimating optimal gain functions and handling non-stationary noise components.
Innovation Solution
A level-dependent maximum noise suppression method is introduced, where the minimum gain is adjusted based on the input signal level, using a level-dependent maximum noise suppression function to control noise suppression effectively, especially for sudden bursts of loud noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-generated harmful factors
If a fixed lower bound is applied to the gain function to minimize perceptual artifacts, then noise modulation and musical tone are reduced, but noise suppression performance is degraded for sudden bursts of loud noise
Solution Approach 1:
The patent applies dynamics by making the lower bound of the gain function adaptive rather than fixed. The lower bound dynamically adjusts based on the input signal level, allowing the system to suppress loud noise effectively while maintaining natural-sounding residual noise. The gain function is defined as G(ω) = max({tilde over (G)}(ω), fLevel(XLevel)), where the lower bound fLevel depends on the input signal level XLevel, enabling dynamic adaptation to different noise conditions.
2Object-generated harmful factors
If the gain function is limited to a lower bound to avoid noise pumping, then perceptual quality is improved, but complete suppression of high-level noise is not achieved
Solution Approach 1:
The patent applies parameter changes by modifying the lower bound parameter based on input signal level. Instead of using a constant lower bound, the system uses a level-dependent lower bound that changes with the noise level. This allows the gain function to achieve complete suppression for loud noise (by reducing the lower bound) while maintaining perceptual quality (by increasing the lower bound for softer noise).
3Adaptability or versatility
If long-term SNR estimation is used to determine the lower bound, then adaptive noise suppression is achieved, but sudden bursts of loud noise are not properly taken into account
Solution Approach 1:
The patent applies preliminary action by using the input signal level XLevel as a direct indicator of current noise conditions before finalizing the gain function. This preliminary measurement of the actual signal level allows the system to immediately respond to sudden bursts of loud noise, rather than waiting for long-term SNR estimation to catch up to rapid changes in noise level.
Data Source
AI summary
A method for level-dependent minimum gain based on a maximum noise suppression function in a voice processing device, the method comprising receiving, by a processor, an input signal comprising noise, determining, by the processor, a level-dependent minimum gain based on level-dependent maximum noise suppression and a level of the input signal and suppressing, by the processor, the noise of the input signal, wherein the noise is suppressed based on the level-dependent minimum gain, wherein the level-dependent maximum noise suppression function provides lower level-dependent minimum gain for higher levels of the input signal and wherein the level of the input signal comprises an amplitude or a power of the input signal.