Speech noise reduction method

TL;DR: In this article, a speech noise reduction method using a traditional signal processing method and a neural network model are combined to complement each other, first, a traditional SPM method is used to process an audio signal, a part of uniform environmental noise such as white noise can be filtered, a subsequent NN model can mainly deal with other noises such as carwhistle and other person sounds, at the same time, a VAD operation is performed on a noisy speech sample, unnecessary signals are removed, and noise reduction can be trained to a certain extent.

Abstract: The invention discloses a speech noise reduction method. A traditional signal processing method and a neural network model are combined to complement each other, firstly, a traditional signal processing method is used to process an audio signal, a part of uniform environmental noise such as white noise can be filtered, a subsequent neural network model can mainly deal with other noise such as carwhistle and other person sounds, at the same time, a VAD operation is performed on a noisy speech sample, unnecessary signals are removed, and the noise reduction can be trained to a certain extent. Atraditional signal processing method has poor scalability, different noise environments need different methods, although a neural network can handle all noise situations, the method is very dependenton a training sample, and the above problem is solved by the speech noise reduction method.