语音处理的方法、装置、电子设备及计算机可读存储介质
By employing short-time Fourier transform, Mel-scale transformation, and subband partitioning, combined with two-dimensional convolution and recurrent network noise reduction models, the problem of excessive computational resource consumption in scenarios such as wireless Bluetooth headsets is solved, achieving efficient speech noise reduction and background human voice suppression.
CN116052705BActive Publication Date: 2026-07-17BESTECHNIC SHANGHAI CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BESTECHNIC SHANGHAI CO LTD
- Filing Date
- 2023-01-16
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
Existing technologies for voice noise reduction in scenarios such as wireless Bluetooth headsets consume excessive computing resources due to the large neural network structure and numerous parameters.
Method used
By employing short-time Fourier transform, Mel-scale transformation, and subband partitioning, combined with a two-dimensional convolutional and recurrent network denoising model, noise reduction is performed on the time-frequency domain features to generate target speech data.
Benefits of technology
It reduces the computational resource consumption of speech denoising processing, while still achieving good denoising results on low-resource platforms, especially effectively suppressing background human voices in dual-talk scenarios.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN116052705B_ABST
Abstract
本申请属于语音处理技术领域，公开了语音处理的方法、装置、电子设备及计算机可读存储介质，该方法包括，对待处理语音数据进行短时傅里叶变换，获得时频域特征；对时频域特征进行划分处理，获得多个初始时频域子带；对各初始时频域子带进行降噪处理，获得多个降噪时频域子带；基于各降噪时频域子带，生成目标语音数据。这样，通过子带划分，降低了语音降噪处理耗费的计算资源。
Need to check novelty before this filing date? Find Prior Art