Acoustic model training method, device, computer equipment and readable storage medium

Abstract

The application provides an acoustic model training method, device, computer equipment and readable storage medium, wherein the method comprises the following steps: obtaining feedback data corresponding to an initial acoustic model during use of the initial acoustic model, wherein the feedback data comprise positive feedback data for characterizing accurate identification results of the initial acoustic model, and negative feedback data for characterizing incorrect identification results of the initial acoustic model; and modifying and training the initial acoustic model based on the feedbackdata and original training data corresponding to the initial acoustic model. By adopting the method, the utilization of the feedback data actually collected by a client as training data can be realized, so that both time and labor costs are saved; in addition, since the feedback data are homologous with the client, the occurrence of biased training can be avoided and the performance of the acoustic model can be improved.

Description

technical field [0001] The present application relates to the technical field of artificial intelligence, and in particular to an acoustic model training method, device, computer equipment and readable storage medium. Background technique [0002] At present, input methods, clients, and smart devices (such as mobile phones, robots, etc.) can provide voice input and recognition functions, and the voice data input by the user is recognized through the voice recognition system to obtain corresponding recognition results. [0003] As one of the components of the speech recognition system, the acoustic model plays a very important role in the accuracy of the recognition results. In the prior art, when the acoustic model is trained, the audio data from a certain source is generally marked by the internal labeling team of the enterprise to obtain the training data, and then the processed training data is added to the acoustic model training, and Improve the performance of the acou...

AI Technical Summary

Problems solved by technology

[0005] The present application proposes an acoustic model training method, device, computer equipment, and readable storage medium, so as to make full use of the feedback data that can actually be collected on the client as training data, saving time and labor costs, and, because the feedback data and The client is of the same origin, which can avoid the occurrence of training bias and improve the performance of the acoustic model. It is used to solve the problem of obtaining the training data of the acoustic model through manual annotation in the prior art, which not only requires a lot of manpower, financial resources, and time for annotation and re- In addition, there is no guarantee that the collected audio data is from the same source as the client data, which may cause technical problems of biased training

Images