REAL-TIME SPEECH RECOGNITION METHOD AND APPARATUS BASED ON TRUNCATED ATTENTION EQUIPMENT AND COMPUTER-READABLE STORAGE MEDIUM

摘要

The present disclosure provides a real-time speech recognition method, apparatus, device, and computer-readable storage medium based on cut attention. The method includes obtaining cleavage information for cutting the feature sequence of the input voice signal, and cleaving the feature sequence into a plurality of subsequences based on the cleavage information. The method further includes obtaining a real-time recognition result through an attention mechanism for each subsequence. An embodiment of the present disclosure introduces cutting information determined according to a voice signal to a traditional attention model, thereby instructing the attention model to perform attention modeling for each cut, so that real-time speech recognition can be implemented as well as A very high recognition accuracy can be guaranteed. In addition, optionally, the dual head structure provided by some embodiments of the present disclosure can ensure that the amount of calculation in the real-time speech recognition process is less, and the multi-level attention structure provided by some embodiments of the present disclosure can further improve the accuracy of speech recognition.