Abstract: The intelligent voice TVM (ticket vending machine) has the function of voice recognition, and passengers can complete the ticket purchase process through voice. However, the metro station features complex environment, noisy background sound and dense crow, environment the conventional voice recognition technology has difficulty in eliminating the noise, holding back a good voice ticket purchase experience for passengers. The microphone array speech enhancement technology, face key point recognition technology and multi-modal information fusion technology of the new intelligent voice TVM are analyzed. The hardware architecture and ticket purchase process of the system are introduced. Application results show that the system can effectively improve the voice signal-to-noise ratio, meet the voice processing requirements of high accuracy and low delay in the complex environment of the station, effectively improve the voice ticket purchase experience of passengers, and finally shorten the average ticket buying time of passengers.