Adaptive Filter for Audio Dereverberation via Overtone Correlation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech dereverberation technologies face challenges in effectively removing reverberation from acoustic signals, especially in non-minimum phase systems and environments with varying noise sources, as they often require access to impulse response measurements and are limited by assumptions about noise characteristics and the need for multiple microphone recordings.
Innovation Solution
An adaptive filtering method that maximizes cross-correlations among spectral overtones to create an inverse filter for removing channel reverberation, independent of optimization methods, which can be implemented in both analog and digital domains, and is applicable to both pre-recorded and live audio signals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If spectral subtraction methods are used to reduce outside noise sources, then noise reduction is achieved, but the methods are not suited for dereverberation because reverberation is time-varying and dependent on the source audio signal
Solution Approach 1:
The patent changes the fundamental parameter being analyzed from amplitude spectrum (used in spectral subtraction) to phase spectrum. By focusing on phase relationships among harmonics rather than amplitude, the method becomes adaptable to reverberation which affects phase more than amplitude, thereby resolving the unsuitability of traditional spectral subtraction for dereverberation
2Reliability
If linear prediction modeling methods are used to optimize speech dereverberation, then speech intelligibility is improved, but the methods are restricted to single speaker sources or require source separation
Solution Approach 1:
The patent segments the audio signal into individual harmonic components and analyzes their phase relationships independently. By examining each harmonic's phase relative to the fundamental frequency, the method can handle multiple speakers simultaneously without requiring source separation, as each speaker's harmonics maintain their characteristic phase relationships
Solution Approach 2:
Instead of modeling the source speech signal directly (as linear prediction does), the patent inverts the approach by modeling the channel effects (reverberation) on the phase spectrum. This inversion allows the method to work with the degraded signal and recover the original by removing channel effects rather than trying to model the source
3Reliability
If inverse filter of the acoustic space is used to remove reverberation, then dereverberation is achieved, but the method requires the acoustic space impulse response to be known in advance and be minimum phase
Solution Approach 1:
The patent makes the system self-sufficient by extracting all necessary information directly from the recorded signal itself. By analyzing the phase relationships among harmonics in the reverberant signal, the method derives the inverse filter characteristics without needing external impulse response measurements or prior knowledge of the acoustic space
Solution Approach 2:
The patent inverts the traditional approach by not trying to directly measure and invert the acoustic impulse response. Instead, it uses the phase spectrum of the recorded signal to infer the inverse filter characteristics, effectively working backwards from the degraded signal to recover the original
4Measurement precision
If statistical model methods are used to estimate reverberant spectrum, then the reverberant spectrum can be estimated, but the model corresponds only to dense reflections in the late reverberant field and not sparse early echoes
Solution Approach 1:
The patent changes the domain of analysis from amplitude spectrum (where statistical models estimate reverberant energy) to phase spectrum. Phase relationships are preserved differently by early and late reflections, allowing the method to capture both sparse early echoes and dense late reverberation through a unified phase-based approach
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach effectively restores the correlations among overtones, improving speech intelligibility by reducing reverberation without relying on specific optimization methods or impulse response measurements, and is suitable for a wide range of audio signals including speech and music.
Implementation Method 1
determining cross-correlations of the overtones of the signal
Implementation Method 2
removing reverberation from the signal by filtering the signal
Data Source
AI summary
Disclosed herein are systems and methods for removing reverberation from signals. The systems and methods can be applicable to audio signals, for example, to voice, musical instrument sounds, and the like. Signals such as the vowel sounds in speech and the sustained portions of many musical instrument sounds can be composed of a fundamental frequency component and a series of harmonically related overtones. The systems and methods can exploit the intrinsically high degree of mutual correlation among the overtones. When such signals are passed through a reverberant channel, the degree of mutual correlation among the partials can be reduced. An inverse channel filter for the removal of reverberation can be found by employing an adaptive filter technique that maximizes the cross-correlation among signal overtones.