Audio Encoder Noise Reduction via Joint LPC Analysis Filter
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech codecs face challenges in jointly optimizing speech enhancement and coding, particularly in reducing background noise and reverberation, due to fundamental differences between enhancement methods and linear predictive coding techniques, leading to suboptimal quality and increased computational complexity.
Innovation Solution
An encoder system that estimates and reduces background noise using a background noise estimator and reducer, followed by linear prediction analysis with a cascade of time-domain filters, enabling real-time processing and minimizing delay by performing noise reduction within the analysis filter in the time domain.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If background noise reduction is applied as a separate pre-processing block before speech coding, then background noise is reduced, but the overall system complexity increases and joint optimization of quality and delay is difficult to achieve
Solution Approach 1:
The patent merges the background noise reduction module with the speech coding module into a unified joint encoder. The analysis filter combines the LPC analysis filter with a noise reduction filter, allowing both speech coding and noise reduction to be performed simultaneously in a single integrated system rather than as separate cascaded stages.
Solution Approach 2:
The analysis filter in the joint encoder performs multiple functions simultaneously: it conducts linear predictive coding analysis for speech representation and applies background noise reduction filtering. This multi-functional design eliminates the need for separate dedicated blocks for each function, reducing overall system complexity while maintaining both speech coding quality and noise reduction effectiveness.
2Object-affected harmful factors
If overlap-add methods with transforms like STFT are used for speech enhancement, then background noise can be reduced, but the fundamental differences make it difficult to merge with CELP codecs and increase computational complexity
Solution Approach 1:
The patent combines the noise reduction filtering directly with the LPC analysis filter in a unified analysis filter structure. Both filtering operations share the same time-domain processing framework, eliminating the need for separate transform-based processing stages and reducing computational overhead compared to cascaded overlap-add methods.
Solution Approach 2:
The patent replaces complex transform-based overlap-add processing with a more efficient time-domain filtering approach. By implementing noise reduction through direct time-domain filtering within the analysis filter, the system achieves comparable or superior noise reduction with lower computational complexity and better real-time performance.
3Ease of manufacture
If cascaded processing is used for speech enhancement and coding, then processing can be modular, but joint perceptual optimization and joint minimization of quantization noise and interference is difficult to achieve
Solution Approach 1:
The patent integrates the noise reduction filter and LPC analysis filter into a single unified analysis filter, eliminating the modular cascaded structure. This integration allows the system to jointly optimize both noise reduction and speech coding quality in a single processing stage, achieving better overall performance than separate modular stages could achieve independently.
4Loss of time
If time-domain filtering is applied within the analysis filter, then real-time processing and reduced delay are achieved, but the complexity of coordinating multiple filters increases
Solution Approach 1:
The patent merges the noise reduction filter and LPC analysis filter into a single integrated analysis filter that operates simultaneously in the time domain. This unified structure eliminates the need for separate filter coordination stages and reduces algorithmic delay by performing both functions in a single pass through the analysis filter.
Data Source
AI summary
It is shown an encoder for encoding an audio signal with reduced background noise using linear predictive coding. The encoder includes a background noise estimator configured to estimate background noise of the audio signal, a background noise reducer configured to generate background noise reduced audio signal by subtracting the estimated background noise of the audio signal from the audio signal, and a predictor configured to subject the audio signal to linear prediction analysis to obtain a first set of linear prediction filter (LPC) coefficients and to subject the background noise reduced audio signal to linear prediction analysis to obtain a second set of linear prediction filter (LPC) coefficients. Furthermore, the encoder includes an analysis filter composed of a cascade of time-domain filters controlled by the obtained first set of LPC coefficients and the obtained second set of LPC coefficients.