Adaptive Noise Estimation for Speech Signals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for processing noisy speech signals, such as VAD-based and MS algorithms, struggle to accurately estimate noise in non-stationary environments, leading to residual noise or speech distortion, and require significant computational resources and memory.
Innovation Solution
A noise estimation method that transforms the noisy speech signal into the frequency domain, calculates a smoothed magnitude spectrum, and uses an adaptive forgetting factor based on an identification ratio to estimate the noise spectrum, allowing for efficient noise estimation and update in noise-like regions without damaging speech components.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If VAD-based or MS algorithms are used to estimate noise in noisy speech signals, then noise estimation can be performed, but accuracy deteriorates in non-stationary environments leading to residual noise or speech distortion
Solution Approach 1:
The patent applies dynamics by making the forgetting factor adaptive rather than fixed. The forgetting factor is dynamically adjusted based on the correlation between consecutive noise estimates, allowing the system to adapt to non-stationary noise environments. When noise varies rapidly, the forgetting factor increases to give more weight to recent observations, while in stationary conditions it decreases to provide smoother estimates.
Solution Approach 2:
The patent changes the parameter of the forgetting factor from a fixed value to a dynamically adjusted value based on noise correlation. This parameter change allows the noise estimation algorithm to adapt its behavior to different environmental conditions, improving both accuracy and reliability in varying noise scenarios.
2Measurement precision
If conventional noise estimation methods are used, then noise can be estimated, but computational resources and memory requirements increase significantly
Solution Approach 1:
The patent extracts only the essential information needed for noise estimation by using a simplified approach that relies on correlation between consecutive estimates rather than comprehensive spectral analysis. This extraction of key features reduces computational complexity while maintaining estimation capability.
Solution Approach 2:
The patent uses a computationally inexpensive forgetting factor mechanism that can be rapidly updated without requiring significant memory or processing resources. This lightweight approach allows continuous noise estimation with minimal computational overhead compared to conventional methods.
3Object-generated harmful factors
If noise estimation is performed aggressively to remove noise, then noise reduction improves, but speech components may be damaged causing distortion
Solution Approach 1:
The patent implements feedback by using the correlation between consecutive noise estimates to adjust the forgetting factor. This feedback mechanism allows the system to monitor the stability of noise estimates and adjust the degree of noise reduction accordingly, preventing excessive filtering that would damage speech components.
Solution Approach 2:
The dynamic adjustment of the forgetting factor based on noise correlation allows the system to modulate the aggressiveness of noise reduction in real-time. When noise is stable, more aggressive reduction is applied; when noise varies rapidly, the system becomes more conservative to protect speech components from distortion.
Data Source
AI summary
A noise estimation method for a noisy speech signal according to an embodiment of the present invention includes the steps of approximating a transformation spectrum by transforming an input noisy speech signal to a frequency domain, calculating a smoothed magnitude spectrum having a decreased difference in a magnitude of the transformation spectrum between neighboring frames, calculating a search spectrum to represent an estimated noise component of the smoothed magnitude spectrum, and estimating a noise spectrum by using a recursive average method using an adaptive forgetting factor defined by using the search spectrum. According to an embodiment of the present invention, the amount of calculation for noise estimation is small, and large-capacity memory is not required. Accordingly, the present invention can be easily implemented in hardware or software. Further, the accuracy of noise estimation can be increase because an adaptive procedure can be performed on each frequency sub-band.