Training method and device for short-speech speaker recognition model

A speaker model and speaker recognition technology, applied in the field of speaker recognition, can solve the problem of low speaker recognition accuracy, and achieve the effects of good performance, high modeling accuracy, and fine modeling scale

Active Publication Date: 2017-11-21
北京灵伴未来科技有限公司
View PDF5 Cites 30 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0007] The present invention provides a speaker model training method for short speech to solve the problem of low accuracy of speaker recognition for short speech in the prior art

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Training method and device for short-speech speaker recognition model
  • Training method and device for short-speech speaker recognition model
  • Training method and device for short-speech speaker recognition model

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0063] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. However, the present invention can be implemented in many other ways different from those described here, and those skilled in the art can make similar extensions without violating the connotation of the present invention, so the present invention is not limited by the specific implementations disclosed below.

[0064] Technical terms related to the present invention:

[0065] 1. "Text-dependent speaker recognition" and "Text-independent speaker recognition"

[0066] Speaker recognition is divided into two types: text-dependent and text-independent according to whether it is related to the text content of the recognized speech. Usually, a text-related task will create a text set, and the user will record training speech according to the specified text during the training phase to establish an accurate text-related speaker model, a...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention discloses a training method for a short-speech speaker recognition model. The method comprises steps of extracting acoustic features from speech data of a training corpus; performing phoneme labeling on the speech data of the training corpus to obtain phoneme labeling results; performing phoneme-related GMM-UBM model training to obtain a phoneme-related GMM model and a UBM-model of a speaker according to the acoustic features and the phoneme labeling results; and based on the phoneme-related GMM model and the UBM model, generating a phoneme-related i-vector total variation matrix for extracting i-vector features for each phoneme; and extracting the phoneme-related i-vector parameters from data of each speaker using the phoneme-related i-vector total variation matrix, and performing dimension reduction to obtain the phoneme-related i-vector features of the speaker. As a speaker model, the present invention creates a phoneme-related i-vector model for the speaker to make the model more accurate and improves the recognition accuracy.

Description

technical field [0001] The invention relates to the field of speaker recognition, in particular to a training method for a speaker recognition model for short speech. The invention also relates to a device for speaker recognition for short speech using a speaker recognition model. Background technique [0002] As one of the main technologies in the field of speech processing, speaker recognition aims to confirm the identity of the speaker, and has broad application prospects in the fields of mobile interaction, identity verification, and audio monitoring. After decades of development, speaker recognition technology has been widely used. [0003] A speaker recognition system usually includes a speaker training stage and a speaker recognition stage. In the speaker training stage, the system first performs mute removal and noise reduction processing on several training voices provided to obtain as pure and effective voice segments as possible, and then extracts the correspond...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Applications(China)
IPC IPC(8): G10L15/02G10L15/06G10L17/02G10L17/04
CPCG10L15/02G10L15/063G10L17/02G10L17/04G10L2015/025G10L2015/0631
Inventor 庞在虎张志平朱风云
Owner 北京灵伴未来科技有限公司
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products