Acoustic Echo Cancellation via Speech-Noise Disentanglement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing acoustic echo cancellation techniques face challenges in efficiently separating speech and noise sources, leading to residual echo suppression issues and a lack of capturing ambient noise, which affects audio communication quality.
Innovation Solution
A method involving trained deep neural networks for source separation and denoising models to factorize the acoustic echo cancellation process, enabling separate transmission of predicted speech and noise signals, improving model training efficiency and communication quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If acoustic echo cancellation uses simultaneous source separation/denoising and speech diarization, then audio quality is improved, but signal processing capacity is overwhelmed
Solution Approach 1:
The patent divides the simultaneous source separation/denoising and speech diarization tasks into separate consecutive processing stages. First, source separation/denoising is performed on the near-end microphone signal to obtain clean speech and noise components. Then, speech diarization is applied to the separated speech signal. This sequential segmentation reduces the computational burden on signal processing capacity while maintaining audio quality through specialized processing for each task.
2Reliability
If acoustic echo cancellation removes all far-end signal effects, then echo suppression is improved, but ambience and noise information is lost
Solution Approach 1:
The patent extracts and separately processes the far-end echo component from the near-end microphone signal using source separation techniques. Instead of completely removing all far-end signal effects, the system identifies and isolates the echo portion while preserving the near-end speech and ambient noise components. This extraction approach enables effective echo suppression while retaining valuable ambience and noise information for natural audio reproduction.
3Productivity
If acoustic echo cancellation processes only speech signals, then processing efficiency is improved, but communication immersion is reduced
Solution Approach 1:
The patent applies different processing qualities to different signal components. Speech signals receive intensive processing with advanced source separation and diarization for high fidelity. Ambient noise and background sounds receive lighter processing to preserve their natural characteristics and spatial information. This local quality differentiation maintains processing efficiency for speech while enhancing communication immersion through preserved environmental audio context.
Data Source
AI summary
The present disclosure relates to an apparatus, that obtains a far-end signal and a near-end microphone signal, determines, based on at least the far-end signal, a far-end speech signal estimate and a far-end noise signal estimate, determines, based on at least the near-end microphone signal, a near-end microphone speech signal estimate and a near-end microphone noise signal estimate, determines, based on at least the far-end speech signal estimate and the near-end microphone speech signal estimate, a predicted near-end speech signal, determines, based on at least the far-end noise signal estimate and the near-end microphone noise signal estimate, a predicted near-end noise signal and outputs at least the predicted near-end speech signal and predicted near-end noise signal.