Voice awakening performance improvement method and device

Abstract

The invention discloses a voice awakening performance improvement method and device. The voice awakening performance improvement method comprises the following steps of detecting an input signal in avoice frame form, and judging whether the voice frame is a suspected voice frame or a non-voice frame; determining an effective voice segment based on the determined suspected voice frame and the non-voice frame, and determining a weighting factor based on the ratio of the suspected voice frame and the non-voice frame in the input signal; inputting the effective voice section into an awakening model, wherein the awakening model is a multi-classification model obtained after supervised depth neural network learning is carried out on collected awakening word positive example samples and non-awakening word negative example samples; and acquiring the output of the adaptive awakening model, carrying out weighted calculation on the output by using the weighting factor, and determining whether the equipment is awakened or not based on output after weighing calculation. According to the scheme provided by the invention, the false awakening rate of a noise scene can be effectively inhibited, and further, the awakening rate for a specific user can be improved in a targeted manner.

Description

technical field [0001] The invention belongs to the technical field of voice wake-up, and in particular relates to a method and a device for improving voice wake-up performance. Background technique [0002] Among the related technologies, there is no example on the market that improves the wake-up rate and reduces false wake-ups. There is a trade-off relationship between wake-up and false wake-up. At present, one of them is chosen for optimization in the market, and the other is optimized on the basis of keeping one basically unchanged. In terms of reducing false wake-ups, it is more representative to add secondary verification; in terms of improving the wake-up rate, it is mainly to lower the wake-up threshold. [0003] In the process of implementing the present application, the inventor found that the following solutions mainly exist in the prior art: [0004] 1. Add a secondary verification module, set a double threshold, and the wake-up engine will make a preliminary ...

AI Technical Summary

Problems solved by technology

[0006] Among them, on the one hand, the secondary verification will increase the power consumption of the device, increase the delay, and affect the user's interactive experience in practice; in addition, there is a risk of crashes on some devices with limited computing resources, which cannot be popularized to those with weak computing power. On the device, the rapid reduction of false wake-ups is also accompanied by a reduction in the wake-up rate

On the other hand, lowering the wake-up threshold can effectively increase the wake-up rate, but at the same time it will increase false wake-ups.

Images