Complex-Domain Speech Denoising With Subband Frequency Balancing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech processing methods result in de-noised speech with poor clarity due to an imbalance of high and low frequency information.
Innovation Solution
A speech processing method involving subband division and aggregation using a pre-trained noise reduction model based on a deep complex convolution recurrent network, which includes an encoding network, decoding network, and long short-term memory network in the complex number domain, to process both amplitude and phase information for accurate noise reduction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If subband division is performed on the first spectrum of the noisy speech in the complex number domain before noise reduction processing, then the clarity of the de-noised speech is improved by effectively processing both high and low frequency information, but the device complexity is increased due to the additional subband division and aggregation operations
Solution Approach 1:
The spectrum of noisy speech is divided into multiple subbands (e.g., low frequency subband and high frequency subband) before noise reduction processing. This segmentation allows different frequency regions to be processed independently, preventing the loss of high frequency information that occurs when processing the entire spectrum as a whole. The subband division is implemented through filtering operations that separate the spectrum into distinct frequency ranges.
Solution Approach 2:
Different noise reduction processing strategies are applied to different subbands according to their specific characteristics. The complex number domain processing is applied uniformly across all subbands to preserve phase information, while the subband-specific filtering allows tailored processing for low and high frequency regions. This local adaptation optimizes the noise reduction effect for each frequency region while maintaining overall speech quality.
2Device complexity
If the first spectrum of noisy speech is directly input into an existing noise reduction model without subband division, then the device complexity is reduced, but the high frequency speech information is severely lost resulting in poor clarity of de-noised speech
Solution Approach 1:
The spectrum is segmented into subbands before processing, which prevents the severe loss of high frequency information that occurs in direct processing. By dividing the spectrum, the high frequency components are preserved in their dedicated subband and can be recovered during the aggregation phase, avoiding the information loss that would occur with direct full-spectrum processing.
Solution Approach 2:
The patent transitions from direct spectrum processing to subband-domain processing, adding a dimensional layer of frequency decomposition. This dimensional change allows the system to maintain high frequency information that would otherwise be lost, as the subband structure provides an additional organizational dimension for preserving and recovering frequency-specific information.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
Disclosed are a speech processing method and apparatus, and an apparatus for processing speech. An embodiment of the method comprises: acquiring a first spectrum of noisy speech in a complex number field; performing sub-band decomposition on the first spectrum to obtain a first sub-band spectrum in the complex number field; processing the first sub-band spectrum on the basis of a pre-trained noise reduction model, so as to obtain a second sub-band spectrum, in the complex number field, of target speech in the noisy speech; performing sub-band restoration on the second sub-band spectrum to obtain a second spectrum in the complex number field; and synthesizing the target speech on the basis of the second spectrum. By means of the embodiment, the problem of high-frequency and low-frequency information being imbalanced is effectively solved, and the clarity of speech after noise reduction is thus improved.