Punctuation predicting, labeling and voice processing method, device, equipment and program product

Abstract

The invention relates to punctuation prediction, labeling and voice processing methods and devices, equipment and a program product. The method comprises the steps of obtaining text feature data after semantic coding is performed on a to-be-processed text; predicting punctuations and punctuation positions in the text to be processed according to the text feature data or according to the text feature data and the identified punctuations and punctuation positions; and outputting the punctuation and the punctuation position of the to-be-processed text. According to the embodiment of the invention, the punctuations and the punctuation positions in the to-be-processed text are predicted according to the text feature data and the identified punctuations and the punctuation positions, and the punctuation prediction processing is not influenced by the length of the input text any more, so that the length of a punctuation prediction structure is greatly reduced, and the effective punctuation prediction rate is greatly reduced.

Description

technical field [0001] The present application relates to a method, device, device and program product for punctuation prediction, labeling, and speech processing, and belongs to the field of computer technology. Background technique [0002] The punctuation prediction system is used to punctuate the text and can be applied in the ASR system. The output text of the original ASR (Automatic Speech Recognition, automatic speech recognition technology) system does not contain punctuation marks, which will greatly affect the readability of the ASR output text and the accuracy of downstream natural language understanding tasks. Adding a punctuation prediction system can solve this problem. [0003] Punctuation prediction systems in the prior art are generally based on two frameworks, one is a sequence labeling framework and the other is a sequence-to-sequence framework. The advantage of the sequence annotation framework is its simple structure, but the disadvantage is that it ca...

AI Technical Summary

Problems solved by technology

The advantage of the sequence tagging framework is its simple structure, but its disadvantage is that it cannot explicitly model the structural relationship of the output punctuation, resulting in a low accuracy rate of punctuation prediction.

The advantage of the sequence-to-sequence framework is that the structural relationship of the output punctuation is explicitly modeled, and the accuracy rate is high. The disadvantage is that the decoder adopts an autoregressive form, and the decoding speed of the decoder will be affected by the length of the input text sequence, resulting in a slower decoding speed.

Images