In this paper， a optical fiber acoustic sensing system based on the Fabry-Perot （F-P） microcavity structure is studied. The sensor head composed of a zirconia （ZrO2） tube with a high melting point and low thermal expansion coefficient bonded to graphene oxide （GO） diaphragm. Test results show that the cavity length of the F-P microcavity is 92.943 μm. Further， the interference contrast of the reflection spectrum is 25 dB. Additionally， the sound sensing performance of the sensor head is tested. Combined with the multiframe comparison method， the automatic recognition of speech leading to silent frames is realized. The Wiener filtering method based on wavelet packet transform is used to denoise the collected signals. Moreover， the signal-to-noise ratio of this method improves by 1.5 dB compared with that of the Wiener filtering method only， thus improving the quality of the speech. The attributes of the entire system include a simple structure， low cost， high practicability， and wide application scenarios.