Method for improving far-field sound pickup of microphone array based on keywords

Abstract

The invention discloses a method for improving far-field sound pickup of a microphone array based on keywords, including the steps: after the keywords are detected by a keyword recognition engine, performing clustering operation on the cached single-frame sound source orientation to obtain the sound source orientation with high reliability, and then calculating the input signal gain of the far-field sound pickup algorithm by using the signal envelope maximum value, and switching the device to the wake-up state, wherein in a wake-up state, the algorithm includes input signal gain adjustment, beam forming, noise suppression, reverberation elimination, AGC, etc, At this time, the AGC can set a small adjustment range to avoid amplifying the low-amplitude signal as background noise, so that when the target keyword is detected, the sound source direction and the amplitude gain of the input signal in the wake-up state can be determined according to the keyword, thereby improving the speech enhancement effect of the far-field sound pick-up of the microphone array.

Description

technical field [0001] The invention relates to the technical field of speech recognition, in particular to a method for improving far-field sound pickup by a microphone array based on keywords. Background technique [0002] In recent years, artificial intelligence has shown explosive growth. Voice interaction is an important field of artificial intelligence, and far-field voice-friendly human-computer interaction has gradually become the dominant way of voice interaction, such as smart speakers, car voice, etc.; microphone array and far-field The high-quality voice signal provided by the voice pickup algorithm for voice interaction is the premise of voice interaction. [0003] The current mainstream far-field pickup method is: the device works in the monitoring state most of the time, and when the keyword recognition engine detects keywords, the device enters the wake-up state; in the wake-up state, the input voice is recognized as various interactive commands; After the i...

AI Technical Summary

Problems solved by technology

[0011] The purpose of the present invention is to overcome the deficiencies in the above-mentioned background technology, and provide a method for improving microphone array far-field sound pickup based on keywords, which can solve the problems caused by unreliable single-frame sound source direction and low input signal amplitude in the microphone array far-field sound pickup algorithm. The effect of speech enhancement is limited. By caching the single-frame direction information and amplitude envelope of keywords, after the keywords are identified, the integrated cache information can obtain reliable sound source direction and appropriate amplitude gain; with reliable With sound source direction and appropriate input signal amplitude, the voice enhancement effect of microphone array far-field pickup is improved, which can improve the far-field voice recognition rate and improve the human hearing effect

Images