Corpus acquisition method and device

Abstract

The invention discloses a corpus acquisition method and device. The method comprises the following steps: acquiring a voice sample; filtering truncated voice and invalid voice in the voice sample to obtain a qualified voice sample; performing multiple voice recognition on the qualified voice sample to obtain a plurality of corresponding voice texts; comparing the plurality of voice texts to obtain a similarity score; if the similarity score is greater than a preset similarity threshold value, taking the voice sample greater than the similarity threshold value as a voice sample to be labeled, and taking the voice text with the longest text content as a voice text to be labeled; and performing manual annotation on the to-be-annotated voice text to obtain an annotated sample. The invention solves the technical problems of long time period and high cost of obtaining voice samples for training in the prior art.

Description

technical field

[0001] The present application relates to the technical field of speech recognition, in particular to a method and device for acquiring corpus. Background technique

[0002] With the rapid development of artificial intelligence, there are more and more data training tasks based on deep learning. In order to achieve better model quality, it is particularly important to obtain high-quality data sets in the early stage. In order to achieve the effect of human communication with the accuracy of human-computer interaction, it is necessary to collect vertical field corpus as a data set for supervised learning of the recognition engine to obtain a high-quality recognition model. In actual project development, voice data collection accounts for one-third of the entire project development cycle. In order to speed up the project development progress, it is necessary to improve the efficiency of data labeling.

[0003] Looking at the voice research and development depa...

Images