Speech Enhancement With Bandwidth Extension for Low-Latency Ear-Wearables
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing DNN-based speech enhancement systems for ear-worn devices face high computational complexity and latency, making them less feasible for real-time applications in hearing aids, especially due to the frequency-dependent SNR of noisy speech and the high risk of distortion in high frequency bands.
Innovation Solution
Implementing a speech enhancement scheme that processes low-frequency bands using reduced bandwidth signals, combined with blind bandwidth extension to synthesize high-frequency components, reducing computational complexity and latency by utilizing low-pass filters and machine-learning algorithms to enhance speech quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If DNN-based speech enhancement is applied to the full bandwidth signal, then speech quality is improved, but computational complexity and latency increase significantly
Solution Approach 1:
The patent segments the speech enhancement task into two parts: (1) enhance only the low-frequency narrowband signal using DNN, and (2) extend the bandwidth to include high frequencies. This segmentation allows the computationally intensive DNN processing to operate on a reduced bandwidth signal, lowering computational complexity while still improving speech quality in the critical low-frequency range.
Solution Approach 2:
The patent extracts and processes only the essential low-frequency components of the speech signal through DNN enhancement, separating them from the high-frequency components. By taking out only the necessary frequency range for enhancement and applying bandwidth extension separately, the system reduces the computational burden on the DNN while maintaining speech quality.
2Manufacturing precision
If DNN-based speech enhancement is applied to the full bandwidth signal, then speech quality is improved, but processing latency increases
Solution Approach 1:
The patent segments the speech enhancement task into two parts: (1) enhance only the low-frequency narrowband signal using DNN, and (2) extend the bandwidth to include high frequencies. This segmentation allows the computationally intensive DNN processing to operate on a reduced bandwidth signal, lowering computational complexity while still improving speech quality in the critical low-frequency range.
Solution Approach 2:
The patent performs preliminary enhancement on the narrowband signal before bandwidth extension. By first enhancing the low-frequency components and then extending the bandwidth, the system avoids the latency of processing the full bandwidth signal through DNN, achieving real-time performance while maintaining speech quality.
3Manufacturing precision
If speech enhancement is applied to high-frequency bands, then speech quality is improved, but the risk of distortion increases due to frequency-dependent SNR
Solution Approach 1:
The patent applies different processing strategies to different frequency bands: DNN-based enhancement is applied only to the low-frequency narrowband signal where SNR is typically better and processing is more reliable, while high-frequency components are handled through bandwidth extension. This local quality approach ensures that enhancement is applied where it is most reliable, minimizing distortion risk.
Data Source
AI summary
An ear-wearable electronic device is operable to apply a low-pass filter to the digitized voice signal to remove a high-frequency component and obtain a low-frequency component. Speech enhancement is applied to the low-frequency component. Blind bandwidth extension is applied to the enhanced low-frequency component to recover or synthesize an estimate of at least part of the high frequency component. An enhanced speech signal is output that is a combination of the enhanced low-frequency component and the bandwidth-extended high frequency component.