Aiming at the problem that current wearable devices lack of analysis on user’s own motion state

a sound collection earphone device composed of sound collection module

bone conduction sound playback module and STM32 processor is designed

and a user motion state offline recognition algorithm（OLRA） based on breathing sound is proposed. The algorithm implements offline identification of four status such as slow walking

jogging

running and ambient sound through initialization

framing

energy detection

wavelet packet denoising

feature parameter extraction

improved K-means clustering

neural network modeling

recognition

and local sample update. The experimental results show that the OLRA algorithm can improve the correct recognition rate of different users’ motion states and reduce the error acceptance rate.