Adaptive SPP Warping for Low SNR Noise Suppression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing noise suppression methods using MMSE-based noise estimation, such as Log-MMSE and its OLSA modification, suffer from increased musical noise in low signal-to-noise ratio situations and over-suppression of weak speech in noisy conditions.
Innovation Solution
An enhanced MMSE determiner is introduced, which modifies the speech presence probability (SPP) using a sigmoid function responsive to real-time signal-to-noise ratio (SNR) to reduce over-attenuation of speech and musical noise, employing a digital signal processor to analyze frequency-domain representations of noisy audio signals and apply adaptive attenuation factors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If OLSA modification of Log-MMSE is used to achieve maximum attenuation, then noise suppression performance is improved, but musical noise increases in low SNR situations
Solution Approach 1:
The patent applies dynamics by making the speech presence probability (SPP) adaptive rather than static. The SPP is modified frame-by-frame based on the external SNR estimate, allowing the system to dynamically adjust attenuation levels. In low SNR conditions, the modification factor reduces attenuation to prevent musical noise, while in high SNR conditions, it allows maximum attenuation for better noise suppression.
Solution Approach 2:
The patent changes the parameter of SPP by introducing an external SNR-based modification factor. This factor adjusts the SPP value from the original Log-MMSE estimator, transforming it from a fixed algorithmic output to a dynamically adjusted parameter that adapts to actual SNR conditions, thereby resolving the contradiction between noise suppression and musical noise generation.
2Reliability
If OLSA modification is applied to maximize attenuation, then noise suppression is improved, but weak speech is over-suppressed in noisy conditions
Solution Approach 1:
The system dynamically adjusts the attenuation level by modifying SPP based on external SNR estimates. When weak speech is present, the external SNR detector identifies the low SNR condition and the modification factor reduces attenuation, preventing over-suppression of weak speech while still providing noise suppression in higher SNR regions.
Solution Approach 2:
The patent implements feedback by using an external SNR estimate to continuously adjust the SPP modification factor. This feedback loop ensures that when weak speech causes the actual SNR to be lower than expected, the system responds by reducing attenuation, thereby preventing information loss while maintaining noise suppression effectiveness.
3Reliability
If MMSE-based noise estimation is used, then noise suppression is achieved, but accuracy deteriorates in low SNR situations
Solution Approach 1:
The patent introduces an external SNR detector as an intermediary that provides accurate SNR estimates independent of the MMSE algorithm's internal estimates. This external detector acts as a mediator that corrects the inaccurate SNR estimates from the MMSE algorithm in low SNR conditions, thereby improving overall accuracy without sacrificing noise suppression capability.
Data Source
AI summary
Acoustic noise in an audio signal is reduced by calculating a speech probability presence (SPP) factor using minimum mean square error (MMSE). The SPP factor, which has a value typically ranging between zero and one, is modified or warped responsive to a value obtained from the evaluation of a sigmoid function, the shape of which is determined by a signal-to-noise ratio (SNR), which is obtained by an evaluation of the signal energy and noise energy output from a microphone over time. The shape and aggressiveness of the sigmoid function is determined using an extrinsically-determined SNR, not determined by the MMSE determination. The extrinsically-determined SNR is obtained from a long term history of previously-determined speech presence probabilities and a long term history of previously-determined noise histories.