Reverberation Removal Filter Using Non-Recursive Coefficient Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for removing reverberation from voice signals, such as those using autoregressive models or inverse filters, often result in decreased accuracy due to model mismatch and filter instability, especially when the microphone is far from the sound source.
Innovation Solution
An acoustic processing apparatus that repeatedly estimates and updates a reverberation removal filter coefficient based on a model adapted to the specific situation, using a reference signal and a non-recursive filter to ensure stability and accuracy, even when the microphone is distant from the sound source.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional autoregressive models or inverse filters are used for reverberation removal, then the processing can be performed with simple structures, but the accuracy of reverberation removal decreases due to model mismatch and filter instability
Solution Approach 1:
The patent changes the fundamental parameters of the filtering approach by switching from unstable inverse filters to stable non-recursive filters, and from autoregressive models to models that represent observation signals as sums of delayed and filtered acoustic signals. This parameter change in the filter type and model structure resolves the contradiction between accuracy and stability.
Solution Approach 2:
Instead of using inverse filters that attempt to invert the reverberation transfer function (which causes instability), the patent inverts the approach by using a stable non-recursive filter with a model that directly represents the relationship between acoustic signals and observation signals. This inversion of the filtering strategy eliminates instability while maintaining accuracy.
2Area of stationary object
If the microphone is placed far from the sound source, then the coverage area increases, but the accuracy of reverberation removal decreases due to increased reverberation and model mismatch
Solution Approach 1:
The patent changes the model parameters to better fit the physical situation of distant microphone placement. By representing the observation signal as a sum of the acoustic signal and a filtered version of the acoustic signal with delay, the model accurately captures the reverberation characteristics even when the microphone is far from the sound source, thus maintaining accuracy while allowing larger coverage area.
3Device complexity
If approximate filters are used to simplify the reverberation removal process, then the device complexity is reduced, but the accuracy of reverberation removal decreases
Solution Approach 1:
The patent changes the filter type from complex approximate inverse filters to simpler non-recursive filters with clearly defined coefficients. The filter coefficients are determined based on the acoustic signal and observation signal without requiring complex inversion processes, thus reducing device complexity while improving accuracy through a more appropriate model representation.
Data Source
AI summary
An acoustic processing apparatus includes a storage, an estimation unit, and a removal unit. The storage stores therein a reference signal indicating a signal obtained by completing removal of reverberation from a first observation signal included in a first processing section. The estimation unit estimates, on the basis of a model representing an observation signal as a signal obtained by adding a signal obtained by applying a reverberation removal filter to an acoustic signal that is input with a delay and the acoustic signal, a filter coefficient of the reverberation removal filter by using a second observation signal and the reference signal. The removal unit determines an output signal indicating a signal obtained by removing reverberation from the second observation signal by using the second observation signal, the reference signal, and the reverberation removal filter having the estimated filter coefficient.