Speech Enhancement And Recognition Algorithm Based On Graphene Microphone

Voice interaction is becoming increasingly indispensable in human-computer interaction.Voice,as the most natural and convenient way for people to issue instructions,makes up for the lack of physical interaction in specific situations.However,as the basis of voice interaction,voice recognition has not been It still faces the problem of being susceptible to noise interference,which limits its application in noisy environments such as industrial production and military accusations.This paper proposes the use of a wearable flexible graphene microphone and a speech enhancement algorithm to achieve accurate speech recognition in a noisy environment,which improves the anti-noise ability of the system in terms of sensors and algorithms,respectively.As a two-dimensional nanomaterial,graphene has extremely low mass and ultra-high mechanical strength,which gives it great advantages in the field of acoustic detection,and its excellent toughness allows it to be combined with flexible materials to make flexible acoustic sensors.However,obtaining excellent graphene is still expensive and difficult to produce on a large scale.In this paper,a low-cost laser-induced graphene preparation method is used to successfully manufacture graphene flexible microphones.This method sacrifices part of the high-frequency sound detection performance,but after the acoustic performance measurement,it far meets the voice signal acquisition requirements of this paper.The graphene flexible microphone can be attached to the surface of the throat to effectively collect weak sound waves and mechanical vibrations.It improves the signal-to-noise ratio at the sensor level and has better anti-noise capabilities.In this paper,three speech command recognition algorithms are designed and implemented on the graphene microphone,which are traditional matching algorithm,1-dimensional convolutional network,and circular convolutional network.Among them,the circular convolutional network has the best performance in speech recognition accuracy and flexibility.excellent.For the speech instruction recognition system in this paper,we designed a lightweight speech enhancement network SEnet,which uses the classic framework of encoder-decoder,adds mechanisms such as skip connection,recurrent unit and gated linear channel,and realizes the model Performance improvement and optimization.In addition,using depthwise separable convolution to further reduce the amount of parameters and computational overhead of the model,SEnet has significantly improved speech enhancement performance compared with the baseline model,and has fewer parameters.Using SEnet for noise reduction of graphene microphones,the accuracy of speech recognition at low signal-to-noise ratios increases significantly,further improving the stability of the speech system.