Bone conduction speech enhancement method based on joint dictionary learning and sparse representation

Abstract

The invention provides a bone conduction speech enhancement method based on joint dictionary learning and sparse representation. In a training stage, in an indoor noise-free environment, a special-shaped double-microphone array composed of a bone conduction microphone and an air conduction microphone is used for synchronously collecting training speech, and a joint training set of the bone conduction speech and the air conduction speech is constructed; and short-time inverse Fourier transform is performed on the training signals of the bone conduction speech and the air conduction speech to obtain a time-frequency spectrum amplitude, and a joint speech dictionary of the bone conduction speech and the air conduction speech is learnt on a time-frequency spectrum. In a detection stage, short-time Fourier transform is performed on the bone conduction speech to obtain a time-frequency spectrum amplitude and a phase; the amplitude is projected on a bone conduction speech sub-dictionary of the joint speech dictionary; the air-guided speech time-frequency spectrum amplitude is reconstructed by using the obtained sparse representation coefficient and the air-guided speech sub-dictionary ofthe joint speech dictionary, two methods are provided for enhancing the time-frequency spectrum of the bone conduction speech, and finally short-time inverse Fourier transform is performed to obtain an enhanced bone conduction speech time-domain signal, so that the speech sharpness is improved.

Description

technical field [0001] The invention relates to the field of single-channel speech enhancement, in particular to a bone conduction speech enhancement method based on joint dictionary learning and sparse representation. Background technique [0002] Voice plays a leading role in people's communication activities. Due to the pollution of environmental noise, the human ear and related smart devices including air conduction microphones (referred to as air conduction microphones) receive noisy voice, and the quality and intelligibility of voice will be significantly reduced, which affects people's subjective Auditory experience and speech recognition rate of smart devices. Speech enhancement technology is the main method to solve this kind of problems. How to restore clean speech from noisy speech has always been a problem that people are trying to solve. The voice received by the air conduction microphone is called air conduction voice for short. [0003] Bone conduction mic...

AI Technical Summary

Problems solved by technology

However, due to factors such as the low-pass performance of human body conduction and the limitation of the sensor technology level, the speech clarity received by the bone conduction microphone is low and sounds dull.

Images