Noise Removal Apparatus Using LPC Feature Vector Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for removing noise from voice signals in noisy environments often misclassify non-static noise as static noise, leading to voice distortion, as they lack effective differentiation between static and non-static noise types.
Innovation Solution
A method and apparatus that determine whether a noise signal is static or non-static by calculating feature vectors, specifically using Linear Prediction Coefficients (LPC), and apply tailored noise removal filters based on energy levels across frequency bands, distinguishing between static and non-static noise signals to accurately remove noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single noise removal method is used for all noise types, then the device complexity is reduced, but the voice distortion increases due to misclassification of non-static noise as static noise
Solution Approach 1:
The patent segments noise removal into two distinct paths: one for static noise and another for non-static noise. The system first determines noise type using LPC feature vectors, then applies appropriate removal methods - frequency spectrum subtraction for static noise and time-domain filtering for non-static noise. This segmentation prevents misclassification and voice distortion while maintaining manageable system complexity through modular design.
Solution Approach 2:
The patent changes the parameter of noise removal approach based on noise type classification. When static noise is detected, frequency spectrum subtraction is applied; when non-static noise is detected, time-domain filtering with adaptive parameters is used. This parameter change strategy allows optimal performance for each noise type without requiring a completely different system architecture.
2Measurement precision
If noise type determination using LPC feature vectors and codebook comparison is implemented, then the noise classification accuracy is improved, but the processing time and computational complexity increase
Solution Approach 1:
The patent performs preliminary action by pre-generating a codebook of LPC feature vectors representing various noise types before actual noise removal. During runtime, the system quickly compares incoming noise LPC vectors against this pre-computed codebook to determine noise type. This preliminary preparation significantly reduces real-time processing time while maintaining high classification accuracy.
3Manufacturing precision
If different noise removal methods are applied based on noise type, then the voice distortion is reduced, but the device complexity and processing steps increase
Solution Approach 1:
The patent introduces an intermediary noise type determination module that acts as a mediator between the input signal and the noise removal processes. This intermediary classifies noise using LPC feature vectors and directs the signal to the appropriate removal path. While this adds a processing step, it prevents the much greater complexity that would result from attempting to handle all noise types with a single complex method.
Data Source
AI summary
A method of removing a noise signal from an input signal, the method including receiving a pure noise signal and an input signal including the noise signal; determining whether the pure noise signal is a static noise signal or a non-static noise signal; and removing the noise signal from the input signal according to whether the noise signal is determined to be the static noise signal or the non-static noise signal.