Time-Frequency Filter Switching for Robust Speech Dereverberation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Dereverberation performance of existing WPE methods is deteriorated in noisy environments and underdetermined conditions due to model errors.
Innovation Solution
A reverberation removal device that applies multiple reverberation prediction filters to an observation signal, switching them according to each time frequency bin to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a single reverberation prediction filter is used in WPE method, then the device complexity is low, but the dereverberation performance deteriorates in noisy environments and underdetermined conditions
Solution Approach 1:
The patent divides the single filter system into multiple segment filters (first reverberation prediction filter and second reverberation prediction filter). Each filter is designed to handle specific characteristics of the observation signal, allowing the system to segment the complex dereverberation task into manageable parts that can be selectively applied
Solution Approach 2:
The patent implements dynamic switching between different reverberation prediction filters based on the characteristics of each time-frequency bin. The switching mechanism dynamically selects which filter to apply (first or second filter) depending on whether the signal exhibits features better handled by one filter or the other, making the system adaptive rather than static
2Measurement precision
If multiple reverberation prediction filters are applied to each time frequency bin, then the measurement precision of dereverberation is improved, but the computational complexity increases
Solution Approach 1:
The patent applies different filter characteristics to different time-frequency bins based on their local signal characteristics. Instead of using a uniform filtering approach across all bins, the system tailors the filter selection to the specific properties of each bin, applying the first filter where appropriate and the second filter where its characteristics are better suited
Solution Approach 2:
The switching mechanism dynamically determines which filter to apply to each time-frequency bin based on real-time signal characteristics. This dynamic adaptation allows the system to optimize dereverberation accuracy for each bin while avoiding the computational burden of applying all possible filters to all bins
Data Source
AI summary
Provided is a reverberation removal device that is highly accurate even in noisy environments and underdetermined conditions. Reverberation is removed by applying a plurality of reverberation prediction filters to an observation signal while switching the plurality of reverberation prediction filters according to each time frequency bin of the observation signal.