Single-channel speech enhancement method and system for neural network sub-band modeling, and storage medium

• Main Authors: LI XIN, ZHOU YANBING, SUN CHONGWU, LIU MING, ZHAO XUEHUA
• Format: Patent
• Language: Chinese; English
• Published: 24.11.2020
• Subjects: ACOUSTICS; MUSICAL INSTRUMENTS; PHYSICS; SPEECH ANALYSIS OR SYNTHESIS; SPEECH OR AUDIO CODING OR DECODING; SPEECH OR VOICE PROCESSING; SPEECH RECOGNITION

The invention provides a single-channel speech enhancement method and system for neural network sub-band modeling and a storage medium, and the method comprises the following steps: 1, collecting a noisy speech signal, and transmitting the noisy speech signal to an logarithm power spectrum extraction module and a Bark cepstrum coefficient extraction module; 2, receiving the noisy voice signal in the step 1 by adopting the logarithm power spectrum extraction module and the Bark cepstrum coefficient extraction module, performing feature extraction on the noisy voice signal by adopting the logarithm power spectrum extraction module and the Bark cepstrum coefficient extraction module, and finally sending the extracted features to a frequency band feature division module; and 3, receiving the features extracted in the step 2 by using a frequency band feature division module, and performing sub-band feature distribution on the extracted features by using the frequency band feature division module. The beneficial ef