Speech feature mapping method and system based on Bhattacharyya distance

Abstract

The invention, which belongs to the technical field of speech recognition / speaker recognition, discloses a speech feature mapping method and system based on Bhattacharyya distance. Features of a speech signal in a complex environment and a clean speech signal are extracted respectively; mapping features are initialized by using a complex feature and a feature mapping formula and GMM models of themapping features and the clean features are established respectively; a minimum Bhattacharyya distance between the two GMM models is estimated iteratively by using an EM algorithm and a final mappingfeature is obtained; and mode matching and identification are carried out on the mapping features and the trained speech signal model in a clean environment. According to the invention, the complex feature mapping is carried out by minimizing the Bhattacharyya distance between the complex feature GMM and the clean feature GMM to obtain mapping features; and mode matching and identification are carried out on the mapping features and the clean model. The mapping features are used for replacing the complex features, so that the speech recognition accuracy is improved effectively.

Description

technical field [0001] The invention belongs to the technical field of speech recognition / speaker recognition, and in particular relates to a speech feature mapping method and system based on Bhattacharyachian distance. Background technique [0002] At present, the existing technologies commonly used in the industry are as follows: [0003] As computer technology promotes the development of human-computer interaction, voice interaction is more and more widely used in real-world scenarios. Voice interaction is the interaction between humans and machines in the form of voice, which is closer to the interaction between humans and more in line with human interaction habits. Voice interaction can enable users to interact with machines more comfortably, make the interaction proceed in a simpler, faster, and more efficient manner, and also make the human-computer interaction process more humanized, highlighting the dominant position of humans in the entire interaction process. V...

AI Technical Summary

Problems solved by technology

[0008] (1) The existing technology only uses the real distribution of the features, and starts mapping without any assumptions or operations on the real distribution of the features. It is difficult to guarantee the mapping direction of the features during the mapping process under this mapping method

[0009] (2) The existing technology does not take more considerations when measuring the distribution of complex features and the distribution of reference features

The Bhattacharyian distance is a good choice for determining the distance between two distributions, which is not exploited by the prior art

[0010] (3) The existing technology does not add prior information when processing complex features

Images