Noise Power Estimation via Cumulative Histogram
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing noise power estimation systems for speech recognition in noisy environments require level-based threshold parameters, which are difficult to set and lead to performance degradation when noise is non-steady-state, lacking robustness against noise environment changes.
Innovation Solution
A noise power estimation system that generates a cumulative histogram weighted by exponential moving average for each frequency spectral component, allowing for noise power estimation without level-based threshold parameters and enhancing robustness against noise environmental changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If MCRA method with fixed threshold parameter is used for noise spectrum estimation, then the system is simple to operate, but it lacks robustness against non-steady-state noise environment changes
Solution Approach 1:
The patent applies dynamics by replacing the fixed threshold parameter with a dynamic cumulative histogram that continuously adapts to changing noise environments. The cumulative histogram is updated recursively using exponential moving average, allowing the noise power estimation to dynamically track non-steady-state noise characteristics without requiring manual threshold adjustment.
Solution Approach 2:
The patent changes the parameter representation from a fixed threshold value to a cumulative histogram distribution. This parameter transformation allows the system to capture the statistical distribution of noise power levels over time, providing robust estimation that adapts to environmental changes while maintaining operational simplicity through automated histogram updating.
2Ease of operation
If level based threshold parameters are used for noise estimation, then the estimation process is straightforward, but it fails to adapt to varying noise conditions
Solution Approach 1:
The cumulative histogram serves as a universal structure that handles multiple noise conditions (steady-state and non-steady-state) with a single estimation framework. Instead of requiring different threshold parameters for different noise types, the histogram-based approach universally adapts to any noise environment by tracking the distribution of power levels across all conditions.
Solution Approach 2:
The system achieves adaptability through dynamic updating of the cumulative histogram using exponential moving average. This allows the estimation process to automatically adjust to varying noise conditions while maintaining the straightforward operation of simply accumulating power level observations without complex parameter tuning.
3Device complexity
If fixed threshold parameters are used for speech enhancement, then the system complexity is reduced, but performance degrades in non-steady-state noise
Solution Approach 1:
The system applies self-service by automatically updating the cumulative histogram and deriving noise power estimates without external parameter adjustment. The exponential moving average mechanism enables the system to self-adapt to changing noise environments, eliminating the need for complex threshold parameter management while maintaining high performance in non-steady-state conditions.
Solution Approach 2:
The patent implements feedback through the recursive updating of the cumulative histogram, where past noise observations continuously inform current estimates. The exponential moving average provides a feedback mechanism that weights recent observations more heavily, allowing the system to respond to environmental changes while maintaining stability, thus improving reliability without increasing complexity.
Data Source
AI summary
A noise power estimation system for estimating noise power of each frequency spectral component includes a cumulative histogram generating section for generating a cumulative histogram for each frequency spectral component of a time series signal, in which the horizontal axis indicates index of power level and the vertical axis indicates cumulative frequency and which is weighted by exponential moving average; and a noise power estimation section for determining an estimated value of noise power for each frequency spectral component of the time series signal based on the cumulative histogram.