Speech recognition method and device, electronic equipment and computer readable storage medium

Abstract

The embodiment of the invention discloses a voice recognition method and device, electronic equipment and a computer readable storage medium, and the method comprises the steps: obtaining to-be-recognized voice data, inputting the to-be-recognized voice data into a first voice recognition model obtained through pre-training, and carrying out the voice recognition, obtaining one or more first voice recognition results and first voice recognition evaluation scores corresponding to the first voice recognition results; in response to detection that a special speech recognition model is preset, inputting the speech data to be recognized into the special speech recognition model to obtain one or more second speech recognition results and corresponding second speech recognition evaluation scores; and mixing and sorting the voice recognition results to obtain a target voice recognition result. According to the technical scheme, the personalized recognition resources can be quickly generated, accurate recognition of personalized contents is realized by calling the personalized recognition resources, and the technical scheme is high in recognition rate, low in calculation amount and low in performance overhead.

Description

technical field [0001] Embodiments of the present disclosure relate to the technical field of speech recognition, and in particular to a speech recognition method, device, electronic equipment, and computer-readable storage medium. Background technique [0002] With the development of science and technology, many smart products are constantly coming out. In order to improve the user experience and enhance their intelligence, many smart products support voice interaction with users, which requires these smart products to have high voice recognition At the same time, considering the needs of personalized voice recognition in practical applications, for example, the operators of smart products have the needs of identifying popular resources and active passwords, and the users of smart products also have the needs of identifying personalized passwords and personal address books. Therefore, smart products are also required to have personalized speech recognition capabilities. Co...

AI Technical Summary

Problems solved by technology

The disadvantages of this scheme are: the stability of the generalization effect is poor, the word segmentation and training process consume too much time, and the model effect obtained by training is prone to training bias and instability; 2. The online recognition text secondary correction scheme, this scheme needs to Customized content and online recognition content are converted into pronunciation sequences in advance, and then determine whether the two pronunciation sequences can be matched successfully, and if the match is successful, try to change the matched sequence fragments into customized recognition content

The disadvantage of this solution is: the recognition result obtained by each speech recognition needs to perform a pronunciation sequence conversion, and then perform a custom pronunciation sequence retrieval. This process is a very large performance for a high-interaction speech recognition system. In addition, the solution does not strengthen the customized content. In the case of noise in the far field, it is easy to have a deviation between the recognition result and the customized pronunciation sequence, which leads to the failure of the subsequent secondary correction process.

Images