The conventional phase-generated carrier (PGC) technology is faced with a high sampling rate and long calculation time during the demodulation process as it adopts a scheme of mixing with a multi-frequency of the carrier. For this reason, improved signal mixing methods are proposed according to the spectral characteristics of speech signals. Signals collected and converted by a photodetector are mixed with the sine and cosine terms of the carrier respectively. After being processed by a band-pass filter, the signals are demodulated by arctangent and differential cross multiplication approaches. The demodulation results of a simple sinusoidal signal prove that compared with the traditional method, the improved demodulation methods require lower sampling rates and shorter calculation time. The differential cross multiplication approach has a faster signal processing speed, whereas the arctangent method has a better demodulation effect and higher noise resistance. The demodulation results of a real voice signal show that the arctangent method has a favorable demodulation effect.