Real-time voice noise reduction method and device based on mask time domain decoder

Abstract

The invention provides a real-time voice noise reduction method and device based on a mask time domain decoder. The method comprises the steps of extracting features from noisy voice through Stft; inputting the extracted features into a pre-trained neural network to obtain a mask; inputting the mask and the noisy voice a time domain decoder to be decoded to obtain enhanced voice, wherein real-time neural network noise reduction based on time domain post-processing is achieved by applying a set of weighting functions (masks) to the time domain decoder to process the noisy voice, the model size is obviously small, the minimum waiting time is shorter, and the method becomes a proper solution for real-time noise reduction of the edge equipment.

Description

technical field [0001] The invention relates to the technical field of speech processing, in particular to a real-time speech noise reduction method and device based on a masked time-domain decoder. Background technique [0002] Speech enhancement refers to the technology of extracting useful speech signals from the noise background and suppressing and reducing noise interference when the speech signal is interfered or even submerged by various noises. Possibly pure original voice. Speech enhancement has a wide range of applications. For speech systems in special environments, some enhancement measures are generally taken to varying degrees. For example, the voice processing of communication in the helicopter cabin and the communication system in the naval ship cabin all require the use of voice enhancement technology. The classic speech enhancement methods are spectral subtraction, Wiener filtering, methods based on statistical models, MCRA minimum recursive averaging met...

AI Technical Summary

Problems solved by technology

This is not apparent from the source reconstruction accuracy even when the ideal clean magnitude spectrum is applied to the mixture

Although phase reconstruction methods can be applied to alleviate this problem, the performance of this method is still suboptimal

Second, decomposing the clean signal from the mixed signal requires a long time window to compute the ISTFT, increasing the minimum delay of the system and limiting its generalizability in real-time, low-latency applications, such as in telecommunications and hearables

Images