Noise Suppression via Speech Likelihood Gain Smoothing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing noise suppressing methods, such as spectral subtraction and minimum mean square error short time spectral amplitude, suffer from estimation errors that lead to residual noise components being dispersed along the time and frequency axes, causing musical noise, and switching between different noise suppression methods can result in unnatural sound changes and distortion.
Innovation Solution
A noise suppressing device and method that estimates a noise spectrum, calculates speech likelihood, and combines suppression gains to smooth the noise suppression process, preventing musical noise and distortion by using a noise estimating unit, speech-likelihood calculating unit, suppression-gain calculating unit, and multiplying unit to produce an output spectrum.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If spectral subtraction or MMSE-STSA method is used to suppress noise in frequency domain, then noise suppression is achieved, but estimation errors cause isolated frequency components to remain dispersed along time and frequency axes, resulting in musical noise
Solution Approach 1:
The patent segments the suppression gain calculation into multiple frequency bands, calculating speech likelihood and suppression gain separately for each band. This allows targeted noise suppression in noise-dominated bands while preserving speech components in speech-dominated bands, preventing the formation of dispersed isolated frequency components that cause musical noise.
Solution Approach 2:
The patent applies different suppression gains to different frequency bands based on local speech likelihood estimates. By adapting the suppression strength to local spectral characteristics, the method suppresses noise where appropriate while preserving speech components, thereby eliminating musical noise caused by uniform or coarse-grained suppression.
2Object-generated harmful factors
If noise suppressing means with coarse frequency resolution is used, then occurrence of isolated frequency components is prevented, but speech component becomes distorted
Solution Approach 1:
The patent divides the frequency spectrum into multiple bands and calculates speech likelihood and suppression gain for each band independently. This fine-grained segmentation allows the system to maintain high frequency resolution for accurate speech component identification while applying suppression at the band level, preventing both isolated frequency components and speech distortion.
Solution Approach 2:
The patent dynamically adjusts suppression gain for each frequency band based on real-time speech likelihood estimation. This dynamic adaptation allows the system to apply strong suppression in noise-dominated bands while applying minimal or no suppression in speech-dominated bands, thereby preventing speech component distortion while eliminating isolated frequency components.
3Manufacturing precision
If noise suppressing means with fine frequency resolution is used, then speech component is less distorted, but isolated frequency components occur causing musical noise in noise-dominant sections
Solution Approach 1:
The patent segments the frequency spectrum into multiple bands and applies different suppression strategies to each band based on speech likelihood. This segmentation allows fine frequency resolution to be maintained for speech preservation while enabling aggressive suppression in noise-dominated bands, thereby eliminating musical noise without distorting speech components.
Solution Approach 2:
The patent changes the suppression parameter (suppression gain) dynamically for each frequency band based on speech likelihood estimates. By adjusting this parameter locally rather than uniformly, the system achieves fine frequency resolution benefits for speech preservation while preventing musical noise through adaptive suppression in noise-dominated bands.
4Object-generated harmful factors
If switching between different noise suppressing methods is performed, then both musical noise and speech distortion are reduced, but drastic changes in output spectrum property occur at switching moments, perceived as unnatural sound
Solution Approach 1:
The patent implements a dynamic suppression gain calculation that continuously adapts to changing spectral conditions through speech likelihood estimation. This dynamic approach eliminates the need for discrete switching between different suppression methods, thereby maintaining output spectrum stability while achieving both musical noise reduction and speech distortion prevention.
Solution Approach 2:
The patent changes suppression parameters continuously based on speech likelihood estimates rather than switching between fixed suppression methods. This continuous parameter adaptation ensures smooth transitions and stability in output spectrum properties while effectively reducing both musical noise and speech distortion.
Data Source
AI summary
There is provided a noise suppressing device for suppress a noise component included in an input signal. The noise suppressing device comprises: a noise estimating unit configured to estimate a noise spectrum based on an input spectrum obtained by performing a frequency analysis on the input signal; a speech-likelihood calculating unit configured to calculate speech-likelihood based on the input spectrum and the noise spectrum; a suppression-gain calculating unit configured to calculate first suppression gain based on the input spectrum and the noise spectrum; a suppression-gain combining unit configured to calculate third suppression gain by combining the first suppression gain and second suppression gain, which is provided as a predetermined constant value or provided by smoothing the first suppression gain, based on the speech-likelihood; and a multiplying unit obtaining an output spectrum by multiplying the input spectrum by the third suppression gain.