A Speech Clarity Enhancement Method Based on Acoustic Feature Transformation

Abstract

The invention discloses a speech clarity enhancement method based on acoustic feature conversion. The method includes a training stage and a use stage; in the training stage, combined with the speaker's noise-fighting vocalization mechanism, the ordinary speech signal and the anti-noise speech under the Lombard effect are used. The signal is used as a data set to train a mapping model with the ability to convert acoustic features. The mapping features include the frequency spectrum slope, fundamental frequency coefficient and energy coefficient of the speech frame. The long short-term memory network is used to learn the feature mapping ability of the frequency spectrum slope, using Bayesian The Gaussian mixture model learns the feature mapping ability of the fundamental frequency coefficient and the energy coefficient; in the use stage, based on the trained feature mapping model and the corresponding pre-processing and post-processing technology, the decoded voice signal of the communication device or interactive device is frame by frame It is converted into a speech signal with anti-noise characteristics to improve the auditory clarity and perceptual intelligibility of the speech signal when it is played in a noisy environment.

Description

technical field [0001] The present invention provides a voice clarity enhancement method based on acoustic feature conversion, especially relates to the fields of artificial intelligence, machine learning, voice signal processing and network communication, and is applicable to systems related to voice communication such as real-time voice communication systems and human-computer interaction systems with equipment. Background technique [0002] With the rapid development of mobile communication technology, relying on powerful mobile communication networks and high-performance call terminals, mobile voice communication has gradually replaced fixed-line communication as the mainstream voice communication method. Relying on the convenience brought by mobile communication, people can carry out voice communication anytime and anywhere. But what follows is that the caller inevitably performs voice communication under the conditions of being surrounded by noisy environments such as...

AI Technical Summary

Problems solved by technology

[0012] The present invention provides a speech clarity enhancement method based on acoustic feature conversion, which solves the problem that when listening to speech in a noisy environment (such as: answering a phone call, listening to an interactive speech of a smart speaker), the original clear speech signal is masked by the environmental noise. Problems that cause loss of aural clarity and perceived intelligibility

Because the traditional digital signal processing method adopts the rigid transfer of spectral energy between the voice frequency bands, without considering the naturalness of the voice, although the clarity of the voice is improved, the naturalness is seriously lacking.

Images