DEEP LEARNING SPEECH EXTRACTION AND NOISE REDUCTION METHOD FUSING SIGNALS OF BONE VIBRATION SENSOR AND MICROPHONE

摘要

A deep learning noise reduction method fusing signals of a bone vibration sensor and a microphone. The method comprises the following steps: S1, a bone vibration sensor and a microphone collecting audio signals to respectively obtain a bone vibration sensor audio signal and a microphone audio signal; S2, inputting the bone vibration sensor audio signal into a high-pass filter module, and performing high-pass filtering; S3, inputting the bone vibration sensor audio signal subjected to high-pass filtering or a signal subjected to frequency band broadening and the microphone audio signal into a deep neural network model; and S4, the deep neural network model obtaining, by means of prediction, speech having been subjected to fusing and noise reduction. In combination with signals of a bone vibration sensor and a traditional microphone, the method uses the high modeling capability of a deep neural network to realize a very high vocal reproduction and an extremely high noise suppression capability, can solve the problem of vocal extraction in a complicated noise scenario, realizes the extraction of a target human voice, reduces interference noise, and can use a single microphone structure to reduce costs. A signal obtained by means of performing frequency band broadening on a bone vibration sensor audio signal can also be directly used as an output.