Speech recognition and automatic error correction method for management instruction

Abstract

The invention discloses a voice recognition and automatic error correction method for a management instruction, and the method comprises the steps: collecting the voice data, used for the management instruction, of a specific management instruction, and carrying out the preprocessing of the voice; extracting voice features; establishing a voice recognition module, and inputting management instruction voice data after feature extraction; outputting a character recognition result corresponding to the management instruction voice to form a text to be corrected; establishing a speech recognition text error correction module, and inputting a to-be-corrected text corresponding to the management instruction speech; outputting the recognition text after the primary error correction; checking the error correctability of each recognition result sequence by taking sentences as units; checking and recognizing the error correctability of the continuously corrected Chinese characters in the text by taking the character as a unit; and outputting the final recognition text after error correction.

Description

technical field [0001] The invention relates to the technical field of speech recognition and speech recognition error correction, in particular to a management instruction-oriented speech recognition and automatic error correction method. Background technique [0002] As one of the key links in the automation of specific industries (such as transportation, public security, network information management, etc.), the automatic speech recognition (ASR) system for managing command speech is improving the automation level and efficiency of specific industries and reducing the work pressure of personnel. play an important role. In the voice communication of management instructions in a specific industry, various management instructions or management intentions of personnel are issued by the management personnel through voice, and the operating machines or personnel perform corresponding operations by recognizing the voice intentions. Admin Command Speech Recognition converts voi...

AI Technical Summary

Problems solved by technology

[0007] c) Mispronunciation, unclear pronunciation, and modal particles such as "啊(a)" and "嘻(ai)" will affect the recognition of management instructions;

[0008] d) Noise during transmission may cause unexpected errors

[0009] However, some methods to improve recognition accuracy, such as using a language model at the ASR decoder to re-score the output n sequences, do not learn from previous error experience, and the performance is not high when the speech environment deteriorates; traditional speech recognition error correction methods Divide error detection and error correction into two parts, and the problem of error propagation and accumulation is serious

Images