Voice breakpoint detection method, device and equipment based on artificial intelligence

Abstract

The embodiment of the invention provides a voice breakpoint detection method, device and equipment based on artificial intelligence, and the method comprises the steps: the semantic integrity detection of candidate results with the probability being higher than a preset threshold value is carried out through a pre-trained semantic integrity model, after it is determined that the candidate resultswith the probabilities higher than the preset threshold value are complete in semantics, natural language understanding is conducted on the candidate results with the probabilities higher than the preset threshold value, and intentions corresponding to the candidate results with the probabilities higher than the preset threshold value are obtained. And finally, a response corresponding to the query statement is obtained according to the candidate result of which the probability is higher than a predetermined threshold and the corresponding intention. Therefore, whether the user speaks or not can be dynamically judged according to the semantic integrity, the real intention of the user can be recognized more accurately, whether the user speaks or not can be accurately judged under the scenesof repeated speaking, sound dragging and the like of the user, and then the user experience can be improved.

Description

technical field [0001] The present application relates to the technical field of speech recognition in artificial intelligence, in particular to a speech breakpoint detection method, device and equipment based on artificial intelligence. Background technique [0002] Automatic Speech Recognition technology (Automatic Speech Recognition; hereinafter referred to as: ASR) is a technology for converting human speech into text. The ASR speech recognition service is often triggered by a wake-up word or a button, and the end point of the speech (Endpoint; hereinafter referred to as: EP) depends on the automatic detection of the ASR. [0003] The EP detection scheme in the existing related art is mainly based on voice activity detection (Voice Activity Detection; hereinafter referred to as: VAD), and there are mainly two kinds of speech breakpoint detection schemes in the existing related art: a detection method based on silence, and a detection method based on prosody and tone. de...

AI Technical Summary

Problems solved by technology

[0005] 1) The effectiveness of the recognition of scenarios such as repeated speech and / or procrastination of the speaker is not high, error-prone, and user experience is unnatural;

[0006] 2) The acoustic model does not understand the content of the user's speech, but only judges based on the audio stream itself, the probability of misjudgment is high, and the deviation is large

Images