Method and system for generating multi-channel noisy speech

Abstract

An embodiment of the present invention provides a method for generating multi-channel noisy speech. The method includes: receiving a near-field single-channel pure voice audio set recorded in a quiet environment by the first recording device for collecting wake-up words; The final direct audio is superimposed to determine the far-field single-channel pure voice audio set in the reverberation environment; according to the phase delay function of the second recording device that collects noise, the far-field single-channel pure voice audio set is simulated as the second recording device for recording A collection of far-field multi-channel voice-only audio; it is mixed to generate far-field multi-channel noisy voice audio in batches. The embodiment of the present invention also provides a system for generating multi-channel noisy speech. The embodiment of the present invention adapts and adjusts the audio recorded by ordinary equipment, which reduces the configuration requirements of the recording equipment in the wake-up word training, and the personnel do not need to go to the recording site to record, which improves the collection efficiency of multi-channel noisy voice.

Description

technical field [0001] The invention relates to the field of wake-up word customization, in particular to a method and system for generating multi-channel noisy speech. Background technique [0002] Wake-up word customization needs to obtain a large amount of noisy voice and audio, and the recording device needs to be placed in a noisy environment. At the same time, people need to speak the wake-up word at a certain distance, and record multi-channel noisy voice data through the recording device. [0003] In the improved version of the wake-up word customization, it is first necessary to place an environmental noise source next to the recording device, the recording device records multi-channel audio data of pure noise, and then records the wake-up word spoken by a person in a quiet environment at a certain distance as a multi-channel The pure voice data, and finally, the multi-channel audio data with pure noise and the multi-channel pure voice data are mixed in a certain wa...

AI Technical Summary

Problems solved by technology

[0007] In order to at least solve the need for certain sound quality and parameters of the noisy voice training wake-up words in the prior art, a large number of recording personnel can only be unified in a specific recording site for recording, making the efficiency of noisy voice collection low. The long collection period makes the wake-up word training less efficient

Images