Voicing-Driven Adaptive Line Enhancer for Periodic Noise Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Single-channel noise reduction methods are inadequate in suppressing short-duration deterministic periodic noise components, such as horn-type sounds and dish clashing, while preserving speech quality, as they rely on stationary noise assumptions and real-valued gain functions, neglecting phase information and failing to adapt effectively to non-stationary noise.
Innovation Solution
A voicing-driven adaptive line enhancer (ALE) with complex-valued filters and pitch-driven adaptation control, which uses voicing signals to set filter coefficients and adjust step-sizes based on pitch frequencies and noise powers, thereby selectively reducing noise without attenuating speech harmonics, and exploits both amplitude and phase information in the frequency domain.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If traditional single-channel noise reduction methods are used, then implementation is simple, but they fail to suppress short-duration deterministic periodic noise components effectively
Solution Approach 1:
The patent implements dynamic adaptation by using a voicing detector to identify voiced speech segments and adjusting the filter coefficients accordingly. The system switches between different filter configurations based on whether the current segment contains voiced speech, allowing it to adapt to changing noise conditions while preserving speech quality. This dynamic approach enables effective suppression of periodic noise components that traditional static methods cannot handle.
Solution Approach 2:
The patent changes key parameters including using complex-valued filter coefficients instead of real-valued, adjusting the filter order dynamically based on voicing detection, and modifying the adaptation step-size according to noise characteristics. These parameter changes enable the system to effectively track and suppress periodic noise components while maintaining speech quality, resolving the contradiction between simplicity and effectiveness.
2Use of energy by moving object
If real-valued gain functions are used, then processing is computationally simple, but phase information is neglected and speech quality deteriorates
Solution Approach 1:
The patent transitions from real-valued processing to complex-valued processing, adding the phase dimension to the filtering operation. By using complex-valued filter coefficients, the system can manipulate both magnitude and phase of the signal components, enabling effective noise suppression while preserving speech quality. This dimensional extension allows the system to exploit phase information that was previously neglected.
3Device complexity
If stationary noise assumptions are made, then processing is simplified, but non-stationary periodic noise cannot be suppressed effectively
Solution Approach 1:
The patent implements feedback mechanisms through voicing detection and adaptive filter coefficient adjustment. The system continuously monitors the input signal to detect voiced speech segments and uses this information to adjust the filter parameters in real-time. This feedback loop enables the system to adapt to non-stationary periodic noise while maintaining manageable processing complexity, as the adaptation is triggered only when needed.
Solution Approach 2:
The patent performs preliminary voicing detection and noise characterization before applying the full noise reduction processing. By identifying voiced speech segments in advance and pre-adjusting filter parameters based on detected noise characteristics, the system prepares optimal processing parameters before the actual noise suppression occurs, enabling effective handling of non-stationary noise without excessive computational complexity.
4Reliability
If adaptive filtering is applied to all frequency components, then noise suppression is maximized, but speech harmonics are attenuated
Solution Approach 1:
The patent applies different filtering strategies to different frequency components and time segments based on voicing detection. During voiced speech segments, the filter adaptation is controlled or suspended to preserve speech harmonics, while during unvoiced segments or noise-dominated periods, aggressive noise suppression is applied. This local differentiation allows the system to maximize noise suppression where safe and preserve speech quality where critical.
Data Source
Figure 1a~1b
Figure 2
Figure 3
AI summary
A signal processor comprising: an input terminal, configured to receive an input-signal; a voicing-terminal, configured to receive a voicing-signal representative of a voiced speech component of the input-signal; an output terminal; a delay block, configured to receive the input-signal and provide a filter-input-signal as a delayed representation of the input-signal; a filter block, configured to: receive the filter-input-signal; and provide a noise-estimate-signal by filtering the filter-input-signal; a combiner block, configured to: receive a combiner-input-signal representative of the input-signal; receive the noise-estimate-signal; and combine the combiner-input-signal with the noise-estimate-signal to provide an output-signal to the output terminal; and a filter-control-block, configured to: receive the voicing-signal; receive signalling representative of the input-signal; and set filter coefficients of the filter block in accordance with the voicing-signal and the input-signal.