This research is dedicated to the optimization of time delay processing for acoustic temperature measurement signals within engine combustion chambers, aiming to significantly enhance the precision and reliability of temperature determinations. A novel signal processing methodology is introduced, integrating wavelet packet decomposition transformation recombination with the isolation forest algorithm. This approach promises to refine the quality of temperature data by effectively mitigating noise interference and extracting pivotal information. Firstly, the acoustic temperature probe was thermally calibrated in a wind tunnel to obtain the data in a high temperature airflow environment. Secondly, the wavelet packet decomposition transformation and reorganization method combined with the isolated forest algorithm was used to filter and reconstruct the temperature data to eliminate noise and extract effective information. At the same time, outliers in the reconstructed data was detected to improve data quality and accuracy. The results of thermal calibration wind tunnel experiments show that the data distribution after signal processing is smoother and more symmetrical, the standard deviation is significantly reduced, and the data is more concentrated on the mean, so as to improve the accuracy and stability of temperature measurement. The research provides an effective technical solution for acoustic temperature measurement.