Multi-Sensor Noise Suppression Using Non-Negative Matrix Factorization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional noise suppression techniques are inadequate for handling non-stationary noise in communication devices, as they rely on stationarity and require prior knowledge of noise environments, limiting their effectiveness in changing noise conditions.
Innovation Solution
A system using multiple sensors (microphones) employs non-negative matrix factorization to estimate noise and speech basis vectors, allowing for real-time adaptation to non-stationary noise environments by modeling noise and speech signals independently, thereby reducing noise distortion and improving speech clarity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional single channel noise suppression techniques (spectral subtraction and Wiener filtering) are used, then the system is simple to implement, but it cannot suppress non-stationary noise effectively because it relies on noise stationarity
Solution Approach 1:
The patent transitions from static noise modeling (assuming stationarity) to dynamic noise modeling by using multiple sensors to capture time-varying noise characteristics. The system continuously updates noise estimates as noise conditions change, making the noise suppression adaptive to non-stationary environments rather than relying on fixed statistical properties.
Solution Approach 2:
The patent moves from single-channel processing to multi-channel processing by incorporating signals from multiple sensors. This additional spatial dimension provides more information about the noise field, enabling the system to distinguish between noise and speech more effectively in non-stationary conditions where temporal statistics alone are insufficient.
2Adaptability or versatility
If SNMF technique is used to suppress non-stationary noise, then the system can handle non-stationary noise, but it requires noise information as a priori knowledge which limits application when noise environment changes
Solution Approach 1:
The system performs self-service by automatically estimating noise characteristics from the multi-sensor inputs without requiring external noise models or a priori knowledge. The multiple sensors enable the system to independently characterize the noise environment and adapt to changing conditions in real-time, eliminating the need for pre-programmed noise profiles.
Solution Approach 2:
The system implements feedback by continuously monitoring the signals from multiple sensors and using this information to update noise estimates dynamically. This closed-loop approach allows the system to respond to changing noise conditions by adjusting its noise model based on current observations rather than relying on fixed prior knowledge.
3Adaptability or versatility
If multiple sensors are used to estimate noise and speech basis vectors, then the system can adapt to changing noise conditions in real-time, but the device complexity increases
Solution Approach 1:
The patent applies segmentation by separating the noise and speech components into distinct basis vectors through non-negative matrix factorization. This decomposition allows the system to process and analyze noise and speech independently, reducing the computational complexity of handling mixed signals from multiple sensors and enabling real-time adaptation.
Data Source
AI summary
Techniques are described herein that suppress noise using multiple sensors (e.g., microphones) of a communication device. Noise modeling (e.g., estimation of noise basis vectors and noise weighting vectors) is performed with respect to a noise signal during operation of a communication device to provide a noise model. The noise model includes noise basis vectors and noise coefficients that represent noise provided by audio sources other than a user of the communication device. Speech modeling (e.g., estimation of speech basis vectors and speech weighting) is performed to provide a speech model. The speech model includes speech basis vectors and speech coefficients that represent speech of the user. A noisy speech signal is processed using the noise basis vectors, the noise coefficients, the speech basis vectors, and the speech coefficients to provide a clean speech signal.