The sound signals of power transformer operation contain extensive information regarding the transformer's operational state. Detecting internal mechanical failures and abnormal states holds significant importance. However, fan noise and other low-frequency noises can interfere with the received acoustic signals, resulting in collected signals that do not accurately represent the transformer's vibration acoustic signals. Current Active Noise Control (ANC) techniques perform better in the low-frequency range. This paper proposes using the FxLMS algorithm to remove low-frequency noise from the acoustic pattern, thereby improving the signal-to-noise ratio. The comparison between the soundprint signals acquired with the turbine off and the noise-reduced soundprint signals verifies the method's feasibility.