Earphone and audio processing method and apparatus therefor, and storage medium

An audio processing method for earphones, comprising:

• acquiring a bone conduction signal and a microphone signal when the earphones are in worn states;
• performing phase adjustment on the bone conduction signal to obtain an adjusted bone conduction signal;
• and
• inputting an audio stream, which includes the adjusted bone conduction signal and the microphone signal, into an audio playing apparatus of the earphones to play the audio stream, so that co-channel interference is generated between the adjusted bone conduction signal in the audio stream and a sound which is transmitted to an ear canal through a user's skeletons,
• wherein the acquiring the bone conduction signal when the earphones are in worn states comprises:
• collecting the bone conduction signal when the earphones are in worn states by a bone conduction sensor, and
• performing noise reduction processing on the bone conduction signal collected by the bone conduction sensor to obtain a noise-reduced bone conduction signal, wherein the performing noise reduction processing on the bone conduction signal collected by the bone conduction sensor comprises acquiring a trained neural network adaptive filter through a cloud server, and
• wherein the acquiring the trained neural network adaptive filter comprises:
• acquiring a training set, wherein the training set comprises a pre-collected microphone signal and a corresponding bone conduction signal before noise reduction and a bone conduction signal after noise reduction, the bone conduction signal before noise reduction is a bone conduction signal collected by the bone conduction sensor, and the bone conduction signal after noise reduction is a bone conduction signal obtained by reducing a self-voice component transmitted through air in the bone conduction signal before noise reduction, and
• training the neural network adaptive filter by using the microphone signal and the bone conduction signal before noise reduction in the training set as input side training data and by using the bone conduction signal after noise reduction in the training set as output side training data, to obtain the trained neural network adaptive filter.