Real-Time Non-Stationary Noise Estimation Using LPC Pattern Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional speech enhancement systems in motor vehicles are inadequate for suppressing non-stationary noise, such as transient noises from vehicle acceleration, traffic, and road bumps, as they rely on retrospective noise detection and assume stationary noise conditions, failing to quickly differentiate noise from speech.
Innovation Solution
A method and apparatus that utilize linear predictive coding (LPC) analysis to create a noise model, update it at the audio signal's frame rate, and calculate metrics of similarity and goodness of fit between higher and lower order LPC coefficients to instantaneously separate noise from speech, allowing for real-time noise suppression.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional retrospective noise detection methods are used, then noise suppression can be achieved for stationary noise, but the system cannot quickly differentiate transient non-stationary noise from speech
Solution Approach 1:
The patent applies dynamics by making the noise estimation adaptive and time-varying through exponential weighting that automatically adjusts between tracking current noise (for non-stationary conditions) and maintaining stability (for stationary conditions). The weighting factor transitions dynamically based on signal characteristics, enabling the system to respond quickly to transient noise while maintaining reliability for stationary noise suppression.
Solution Approach 2:
The patent changes the parameter of noise estimation by introducing a frequency-dependent exponential weighting mechanism that modifies how past noise samples are combined. The weighting parameter varies with frequency and signal conditions, allowing the system to differentiate between speech and transient noise by detecting abrupt parameter changes in the noise spectrum that conventional methods miss.
2Stability of the object's composition
If slow temporal smoothing is used to estimate noise power spectrum density, then stability is improved, but the system cannot respond quickly to sudden non-stationary noise
Solution Approach 1:
The system dynamically adjusts the effective smoothing time constant based on signal conditions. During stationary noise periods, strong smoothing provides stability. When transient noise or speech occurs, the exponential weighting naturally reduces effective smoothing, enabling fast response. This dynamic behavior resolves the contradiction between stability and speed.
Solution Approach 2:
The patent implements feedback by continuously monitoring the current frame's signal characteristics and using this information to weight the contribution of past noise estimates. The feedback loop detects when noise characteristics change abruptly and automatically reduces reliance on historical data, enabling the system to maintain stability during stationary conditions while responding quickly to changes.
3Measurement precision
If higher order LPC analysis is used for speech modeling, then speech representation accuracy is improved, but noise components cannot be effectively separated
Solution Approach 1:
The patent segments the LPC analysis into two distinct components: a high-order LPC model for capturing speech structure and a low-order LPC model for representing noise characteristics. By separating these functions into different model orders, the system achieves both accurate speech representation and effective noise separation without requiring a single overly complex model.
Solution Approach 2:
The patent applies local quality by using different LPC model orders for different signal components. High-order LPC is applied locally to speech segments where detailed spectral representation is needed, while low-order LPC is applied locally to noise segments where simple spectral shaping suffices. This localized application of different model complexities optimizes both accuracy and separability.
Data Source
AI summary
Speech in a motor vehicle is improved by suppressing transient, “non-stationary” noise using pattern matching. Pre-stored sets of linear predictive coefficients are compared to LPC coefficients of a noise signal. The pre-stored LPC coefficient set that is “closest” to an LPC coefficient set representing a signal comprising speech and noise is considered to be noise.