Automatic labeling method for Yi language voice data

Abstract

The invention provides an automatic labeling method for Yi language voice data. The automatic labeling method comprises the following steps: 1, respectively processing Yi language samples into Yi language text samples and corresponding Yi language voice samples; performing shared representation on the Yi language text sample in a language layer to obtain a language layer shared identification sample; performing shared representation on the Yi language voice sample in an acoustic layer to obtain an acoustic layer shared identification sample; 2, performing alignment fusion on the language layer shared identification sample and the acoustic layer shared identification sample to complete preprocessing so as to obtain Yi language sample data after alignment fusion of the language layer and the acoustic layer; 3, performing annotation parameter calculation on the aligned and fused Yi language sample data by utilizing an extended Gaussian mixture model and a hidden Markov model GMM-HMM; and 4, taking the annotation parameter calculation result as a basis for annotation, and finally finishing automatic annotation of the Yi language sample data. According to the method, the accuracy of automatic labeling can be improved.

Description

technical field [0001] The invention relates to the technical field of intelligent recognition, in particular to an automatic labeling method for Yi speech data. Background technique [0002] The Yi nationality is one of the most populous ethnic groups in southwestern my country, with a population of about 8.7 million, of which 5.1 million are in Yunnan alone. The Yi language belongs to the Yi branch of the Tibeto-Burman group of the Sino-Tibetan language family. It covers a wide geographical distribution, with many dialects, sub-dialects and native languages. It is divided into six major dialect regions: eastern, southern, western, northern, southeastern and central. The use of Yi language varies from place to place, and more than half of the residents in most areas where the Yi nationality gathers do not understand or basically understand Chinese. The Yi language has many characteristics at the acoustic level (phonemes, accents, consonants, etc.) and linguistics (vocabula...

AI Technical Summary

Problems solved by technology

[0004] The existing data labeling methods mainly include: first, the artificial voice data labeling method, which is costly and the quality of the labeling is uneven, which will affect the effect of later data use; second, the automatic labeling method in the field of images, etc. The automatic labeling method of image data has achieved great success in the field, but it does not have the versatility to be extended to other fields

Third, the automatic voice data labeling method based on single-modal features uses existing voice recognition technology principles to achieve labeling, but because only a certain aspect of feature factors is considered, the accuracy of data labeling is low

Images