Voice detection method based on multiple sound areas, related device and storage medium

Abstract

The invention discloses a voice detection method based on multiple sound areas, the method is applied to the field of artificial intelligence, and the voice detection method provided by the inventioncomprises the following steps: obtaining sound area information corresponding to each sound area in N sound areas; generating a control signal corresponding to each sound area according to the sound area information corresponding to each sound area; processing the voice input signal corresponding to each sound area by adopting the control signal corresponding to each sound area to obtain a voice output signal corresponding to each sound area; and generating a voice detection result according to the voice output signal corresponding to each sound area. The invention further discloses a voice detection device and a storage medium. According to the invention, the voice signals from different directions can be processed in parallel based on the plurality of sound areas, and the voice signals in different directions are reserved or suppressed through the control signals in a multi-sound-source scene, so that the voice of each user can be separated and enhanced in real time, and the accuracyof voice detection is improved.

Description

technical field [0001] The present application relates to the field of artificial intelligence, and in particular to a multi-sound region-based speech detection method, related devices and storage media. Background technique [0002] With the wide application of far-field voice in people's daily life, in the multi-sound source (or multi-user) scenario, perform voice activity detection (VAD), separation, enhancement, and recognition for each possible sound source It has become a bottleneck for many intelligent voice products to improve their voice interaction performance. [0003] In the traditional scheme, a monophonic pre-processing system based on the main speaker detection algorithm is designed. The pre-processing system generally uses azimuth estimation combined with signal strength estimation, or azimuth estimation combined with spatial spectrum estimation. Estimate the speaker with the strongest signal energy (that is, the signal energy reaching the microphone array) ...

AI Technical Summary

Problems solved by technology

[0004] However, when there are multiple speakers in the environment, judging the main speaker only by the arrival signal strength is flawed, because the main speaker may be farther away from the microphone array than the interfering speaker. Although the volume of the main speaker may be greater than that of the interfering speaker, the propagation loss of the speech signal in the space is greater, so the signal strength reaching the microphone array may be smaller instead, resulting in poorer effect on subsequent speech processing

Images