Telephone robot speech recognition result correction method based on deep learning

Abstract

The invention discloses a telephone robot speech recognition result correction method based on deep learning. The correction method comprises the following steps: obtaining a pinyin sentence text dataset Sp and a correct sentence text sample set Sc based on a historical speech data set; establishing a correction model by adopting deep learning; establishing the correction model comprises an encoder part construction based on a multi-head attention model and a feedforward neural network and a decoder part construction based on two stacked multi-head attention models and a feedforward neural network; training the established correction model based on the correct sentence text sample set Sc; and inputting a speech recognition result to be corrected into the trained correction model after being processed by a vectorization procedure to obtain a corrected text. The telephone robot speech recognition result correction method based on deep learning fully utilizes historical recording data resources, trains a speech recognition result correction model, and efficiently recognizes and corrects the speech in an unquiet environment and under the conditions of low speech recognition precisionsuch as a plurality of different speaking modes, pronunciation accuracy, sound receiving capability and the like.

Description

technical field [0001] The invention belongs to the technical field of speech recognition, and in particular relates to a method for correcting speech recognition results of a telephone robot based on deep learning. Background technique [0002] With the formation of a customer-oriented market and the popularization of telephones, more and more companies have begun to use outbound calling systems to expand and maintain customers and increase corporate benefits. Establish a good communication bridge with customers through the outbound call system, understand customer conditions, opinions and needs, actively publicize the company's new policies, new discounts, and recommend new businesses to customers, and timely provide new customers with information on service attitude, product quality, usage, etc. Conduct return visits to achieve the purpose of retaining customers and expanding the number of customers during the process of outbound return visits. [0003] Traditional outbo...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to solve the problem that the speech recognition accuracy drops sharply and the recognition results are inaccurate in the current speech recognition technology for the environment that is not quiet and many different ways of speaking, pronunciation accuracy, and sound collection ability. A method for correcting speech recognition results of telephone robots based on deep learning

Images