Target voice extraction method based on double-microphone array

Abstract

The invention discloses a target voice extraction method based on a double-microphone array. The method comprises the following steps: S1, converting time domain signals received by two microphones into frequency domain signals; S2, performing frequency domain blind source separation algorithm processing on the frequency domain signals; processing to obtain estimation signals of the target voice signal and the noise signal; S3, solving respective voice activity factors and power spectrum values by using the two estimation signals, and immediately caching the voice activity factors and the power spectrum values in a data cache region; and S4, instantly checking whether a wake-up moment exists or not, comprehensively comparing the ratio of the average power spectrum values of the two estimation signals with the difference value of the average voice activity factors at the wake-up moment, and selecting a target voice from the estimation signals. The voice activity factor value and the power spectrum value are compared, the target voice signal is extracted, compared with a traditional method, DOA does not need to be used for providing target sound source prior information, a deep learning method does not need to be used for training mass data extraction, and the requirement for system resources is lowered.

Description

technical field [0001] The invention belongs to the technical field of artificial intelligence, relates to speech extraction technology, and in particular relates to a target speech extraction method based on a dual-microphone array. Background technique [0002] In the field of artificial intelligence technology, speech enhancement and speech recognition have always been hot topics for experts, scholars and the market of speech interactive products. Among them, the dual-microphone array has become the main research object of technicians due to its obvious advantages. The reason is that compared with a single microphone, the dual-microphone array has great advantages in non-stationary noise processing and long-distance sound pickup; and compared with the multi-microphone array, the dual-microphone array greatly simplifies the hardware of voice interaction products. Design scheme and complexity of voice front-end algorithm processing. Therefore, dual-microphone arrays are w...

AI Technical Summary

Problems solved by technology

However, for dual microphones, due to the small number of microphones and weak spatial directivity, the DOA estimation error is large, so that the method using DOA will distort the signal spectrum when the target speech is extracted or even fail to extract the target speech.

However, the deep learning method has high requirements on the performance and resources of the embedded system, and requires a large amount of data to train the model, and the generalization performance will be limited when applied.

Images