Adaptable Post-Filter for Beamforming Signal Enhancement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current noise reduction methods in speech signal processing, particularly in two-way speech communication and speech recognition, are inadequate in handling non-stationary noise and fail to significantly improve signal intelligibility due to limitations in beamforming and post-filtering techniques, leading to insufficient signal-to-noise ratio enhancement.
Innovation Solution
A method that combines beamforming with adaptable post-filtering using previously learned filter weights, where microphone signals are processed by a beamformer and then filtered using a post-filtering mechanism with weights adjusted by a non-linear mapping process, such as a neural network or code book system, to enhance the signal quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If beamforming is used to spatially filter signals, then signal-to-noise ratio is improved, but noise suppression remains limited and frequency-dependent
Solution Approach 1:
The patent combines beamforming with post-filtering to create a hybrid noise reduction system. The beamformer provides spatial filtering and initial SNR improvement, while the post-filter supplements this with frequency-dependent noise suppression, achieving better overall performance than either method alone.
Solution Approach 2:
The post-filter uses adaptive filter weights that are continuously adjusted based on the speech activity detection and noise characteristics. This dynamic adaptation allows the system to optimize noise suppression in real-time while preserving speech quality, overcoming the static limitations of traditional beamforming.
2Object-affected harmful factors
If post-filters with time-dependent spectral weighting are used, then noise reduction is improved, but determination of optimal weights becomes complex and computationally intensive
Solution Approach 1:
The system employs speech activity detection mechanisms that automatically identify speech segments and adapt filter weights accordingly. The speech activity detector monitors the signal and triggers appropriate filtering strategies, enabling the system to self-adjust without complex external control mechanisms.
Solution Approach 2:
The patent changes the operational parameters of the post-filter based on detected speech activity. During speech segments, the filter applies aggressive noise suppression; during non-speech segments, it reduces filtering to preserve background sounds. This parameter adaptation simplifies the overall system by using simple thresholds and gain adjustments rather than complex optimization algorithms.
3Object-affected harmful factors
If spectral components affected by noise are damped, then noise reduction is achieved, but speech signal quality deteriorates and intelligibility is not sufficiently improved
Solution Approach 1:
The system uses speech activity detection as a feedback mechanism to monitor the quality of processed speech signals. Based on this feedback, the post-filter dynamically adjusts its filtering strength to maintain speech intelligibility while reducing noise. This closed-loop approach prevents excessive damping that would degrade speech quality.
Solution Approach 2:
The post-filter applies different filtering characteristics to different frequency bands and time segments based on local signal characteristics. Rather than uniformly damping all spectral components, it selectively suppresses noise in specific frequency regions where speech energy is low, thereby preserving speech quality while achieving noise reduction.
Data Source
Figure 1
Figure 2
AI summary
The present invention relates to a method for speech signal processing, comprising detecting a speech signal by more than one microphone to obtain microphone signals (x1, x2); processing the microphone signals (x1, x2) by a beamforming means (2) to obtain a beamformed signal (XBF); post-filtering the beamformed signal (XBF) by a post-filtering means (6) comprising adaptable filter weights to obtain an enhanced beamformed signal (XP) and adapting the filter weights of the post-filtering means (6) by means of previously learned filter weights.