Broadband speech spectrum inclination characteristic parameter reconstruction method for speech intelligibility enhancement

Abstract

The invention provides a broadband speech spectrum inclination characteristic parameter reconstruction method for speech intelligibility enhancement, comprising a training stage and a use stage of a spectrum inclination reconstruction network based on a cyclic neural network, wherein a training stage establishes a speech data set, and pretreats speech data in the data set; Input the preprocessed narrowband speech data, perform short-time Fourier transform to obtain narrowband speech spectrum, logarithmize the spectrum information to obtain logarithmic amplitude spectrum; Input the preprocessedwideband speech data, extract the parameters of all-pole model of wideband speech signal spectral inclination, and transform them into linear spectral pair parameters. The spectral tilt reconstruction network is trained and used to reconstruct the all-pole model parameters of the spectral tilt of wideband speech. The invention reconstructs the broadband speech signal spectrum inclination parameter according to the narrowband speech signal, and is suitable for all speech intelligibility enhancement systems based on the spectrum inclination characteristic, and can adapt to multilingual and multimodal speech signals.

Description

technical field [0001] The present invention provides a method for rebuilding characteristic parameters of wideband speech spectrum gradient for speech clarity enhancement, relates to the field of speech signal processing and communication technology, is applicable to all speech clarity enhancement systems based on spectrum gradient features, and can Adapt to multilingual and multimodal voice signals. Background technique [0002] Since the 21st century, mobile communication technology has developed rapidly, and mobile communication devices such as mobile phones have become popular rapidly. With the convenience brought by mobile phones, people can use mobile communication devices for real-time voice communication anytime and anywhere; with this convenience, people inevitably talk in diverse noisy environments such as stations, restaurants, factories, etc., and noise in noisy environments Seriously degraded voice call quality. [0003] The voice communication process can be...

AI Technical Summary

Problems solved by technology

[0011] The present invention provides a reconstruction method for characteristic parameters of wideband speech spectrum inclination for speech intelligibility enhancement, which solves the problem that due to the lack of acoustic features of narrowband speech signals, the directly calculated spectral gradient parameters have larger errors than wideband speech signals. The problem that the speech clarity enhancement system cannot obtain accurate spectrum slope information and the enhancement effect is severely reduced

Images