Multi-Channel Kalman Gain Sharing for Acoustic Howling Suppression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing acoustic howling suppression methods, such as gain control, notch filters, and adaptive feedback cancellation, are inadequate in scenarios requiring high acoustic amplification, distort the target sound, or introduce unexpected frequencies, and deep learning-based methods suffer from mismatch issues between training and inference stages.
Innovation Solution
Implementing a deep learning approach called Deep AHS with teacher-forced learning and a Kalman filter to suppress acoustic howling by transforming the recurrent suppression process into an instantaneous speech separation process, using attention-based recurrent neural networks and shared parameter estimation across multiple channels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If gain control is used to reduce amplifier volume, then howling is suppressed, but acoustic amplification capability is reduced
Solution Approach 1:
The audio signal is segmented into different frequency components using Fast Fourier Transform (FFT). The system identifies specific frequency bands where howling occurs and applies gain reduction only to those frequencies, while maintaining full gain for other frequencies. This frequency-domain segmentation allows howling suppression without compromising overall acoustic amplification capability.
Solution Approach 2:
The system applies different processing characteristics to different frequency regions. Notch filters are applied locally at specific howling frequencies, while the rest of the frequency spectrum maintains high gain for amplification. This local quality approach ensures that harmfully suppressed only where necessary, preserving amplification power elsewhere.
2Object-affected harmful factors
If notch filters are used to attenuate howling, then howling frequencies are suppressed, but target sound is distorted
Solution Approach 1:
The notch filter parameters (center frequency, bandwidth, and depth) are dynamically adjusted based on real-time howling detection. The system continuously monitors the audio spectrum and adapts the filter characteristics to match the actual howling frequencies and their evolution over time. This dynamic adaptation prevents fixed-filter distortion of target sounds while maintaining effective howling suppression.
Solution Approach 2:
The system employs feedback mechanisms where the output of the notch filter is monitored and used to further refine filter parameters. The error signal between the original and filtered signals is analyzed to detect residual howling and adjust the filter accordingly, ensuring minimal impact on target sound while maintaining howling suppression effectiveness.
3Object-affected harmful factors
If adaptive feedback cancellation is used to estimate acoustic path, then howling is suppressed, but speech quality is distorted due to de-correlation techniques
Solution Approach 1:
The system introduces an intermediary reference signal that correlates with both the target speech and the feedback path. By using this intermediary signal as a basis for adaptive filtering, the system can estimate the acoustic path without requiring de-correlation techniques that distort speech. The intermediary acts as a bridge that maintains speech integrity while enabling accurate feedback cancellation.
Solution Approach 2:
The patent replaces traditional mechanical/acoustic de-correlation techniques with signal processing-based methods in the frequency domain. Instead of manipulating the physical signal characteristics to achieve de-correlation, the system uses computational approaches (FFT-based filtering and adaptive algorithms) that preserve speech quality while achieving howling suppression through mathematical operations.
4Manufacturing precision
If multi-channel microphones are used to improve speech quality, then user experience is enhanced, but howling suppression complexity increases
Solution Approach 1:
The system merges the processing of multiple microphone channels by combining their signals in the frequency domain before applying the Kalman filter. Instead of independently processing each channel, the multi-channel signals are integrated into a unified processing framework that shares computational resources and parameters, reducing overall complexity while maintaining the speech quality benefits of multiple microphones.
Solution Approach 2:
The Kalman filter implementation serves multiple functions simultaneously: it performs howling detection, howling suppression, and acoustic path estimation across all channels. This multi-functional approach eliminates the need for separate processing chains for each task, reducing system complexity while maintaining effective howling suppression and speech quality enhancement across the multi-channel setup.
Data Source
AI summary
A method and apparatus comprising computer code configured to cause a processor or processors to obtain an audio signal from a channel of at least one microphone of a plurality of microphones, estimate a Kalman gain based on the audio signal, share the Kalman gain to a plurality of channels of other ones of the plurality of microphones, and output an AHS signal from the channel and the plurality of channels, wherein the AHS signal is a version of the audio signal in which acoustic howling noise of the audio signal is suppressed and target audio of the audio signal is sustained.