Method for determining custom wake-up word and device for determining custom wake-up word

Abstract

The invention discloses a method for determining a self-defined wake-up word, which includes: receiving a first user instruction; determining custom content according to the first user instruction; evaluating the self-defined wake-up word; and determining the self-defined wake-up word according to the evaluation result. The present invention also discloses a device for determining a self-defined wake-up word. The method and device provided by the present invention can realize the self-definition of the wake-up word, and can obtain more accurate, high wake-up rate, and low false wake-up rate. Customize the wake-up word, and the process of generating the wake-up word is more efficient and faster. At the same time, because it can well guarantee the quality of the wake-up word in the custom part, it improves the playability of the voice product and greatly improves the user experience of the entire product.

Description

technical field [0001] The present invention relates to the technical field of voice interaction, in particular to a method for determining a custom wake-up word and a device for determining the custom wake-up word. Background technique [0002] With the increasing of voice interaction technology, there are two mainstream voice wake-up models at present, one is based on language model, the other is based on no language model, the wake-up model based on language model includes acoustic model and language model , needs to be verified by two models. Although the accuracy of the verification is high and the utilization rate and availability of the wake-up word are high, it requires a huge amount of calculation, so the processing process of this model is slow and inefficient. For the wake-up model based on the non-linguistic model, only acoustic verification is performed, the calculation amount is small, and the processing speed is fast, but it is more troublesome when the user a...

AI Technical Summary

Problems solved by technology

[0002] With the increasing of voice interaction technology, there are two mainstream voice wake-up models at present, one is based on language model, the other is based on no language model, the wake-up model based on language model includes acoustic model and language model , needs to go through the verification process of two models. Although the accuracy of the verification is high, the utilization rate and availability of the wake-up words obtained are high, but it requires a huge amount of calculation, so the processing process of this model is slow and inefficient.

For the wake-up model based on the language-free model, only the acoustic verification is performed, the calculation amount is small, and the processing speed is fast, but it is more troublesome when the user applies the model to customize the wake-up word, and the accuracy and usability will be reduced.

Images