Method, device and system for speech recognition of AR/MR equipment

Abstract

The invention discloses a method, device and system for speech recognition of AR / MR equipment. The method comprises the following steps: displaying a projection screen and a virtual screen in a virtual view field of AR / MR equipment, wherein the projection screen is formed by projecting a display screen of intelligent handheld equipment into the virtual view field; displaying a voice recognition function interface on a display screen of the intelligent handheld device; and performing voice input and voice recognition according to the operation of the user on the voice recognition function interface of the intelligent handheld device, displaying a plurality of recognized candidate results on a virtual screen, and selecting and confirming the plurality of candidate results. According to the invention, the AR / MR is utilized to display a plurality of candidate results of voice recognition more conveniently and more clearly, and an application interface of the intelligent handheld device is not affected; and touch sliding is carried out through the voice recognition function interface, candidate results can be more conveniently and visually selected, the operation convenience is improved, and operation without leaving the device by hand and single-hand operation can be achieved.

Description

technical field [0001] The present invention relates to the technical field of speech recognition, in particular to a method, device and system for speech recognition of AR / MR equipment. Background technique [0002] Speech Recognition (Speech Recognition) technology, also known as Automatic Speech Recognition (ASR), its goal is to convert the content (Content) in the voice into computer-readable input, such as keystrokes, binary codes, or character sequences ( text), and operate accordingly. [0003] The mainstream technology of speech recognition has been converted from Hidden Markov Model (HMM) to speech recognition system based on Deep Neural Network (DNN). In speech recognition tasks, an acoustic model (Acoustic Model) and a language model (Language Model) are generally required. The current speech recognition is gradually turning to the framework of WFST (weighted finaite-state transducer, weighted finite state transducer) + deep neural network. [0004] During spee...

AI Technical Summary

Problems solved by technology

[0011] 1. If there are multiple candidates for the view (selection list), the main problem is that when there are many candidates, the interface of the original application will be blocked; at the same time, if there are many candidates, the screen may not fit or the font may be too small; and, Sometimes need another hand to help click

[0012] 2. If it is in the form of pure voice prompts, it is not easy for users to remember the candidates, and it is generally inconvenient to listen to them repeatedly. Too many candidates are easy for users to forget

Images