Selecting sound source signals in acoustic temperature measurement technology is particularly important and is a key to improving the accuracy of acoustic temperature measurements. Pseudorandom, linearly swept, and pulsed signals are selected as research objects for this study. The decay of the three signals as they propagate through a loose coal, the frequency band and the period of the linearly swept signal are analyzed. The results showed that the attenuation coefficient of the linear sweep signal is less sensitive to temperature, and the attenuation coefficient of the low-frequency sound wave is smaller. When the lower limit of the frequency band of the sweep signal is closer to the upper limit of the frequency band, the main peak value of the cross-correlation coefficient is smaller. The main peak of the cross-correlation curve of low-band narrowband signals is greater than that of high-band narrowband signals. The cross-correlation effect is more significant for the 600–1000 Hz sweep signal. Moreover, the strongest effect is determined when the period is 0.1 s. Finally, the temperature of anthracite loose coal was measured according to the optimal signal. The maximum absolute error in the measurement results was 1.23 °C, and the maximum relative error was 4.22%.