PURPOSE: To accurately and efficiently enable a compression processing of a sound signal by using a predictive coefficient obtained from adaptive prediction performed in a voice coding processing after converting into a digital signal and deciding the presence of a conversation voice.

CONSTITUTION: An ADPCM code converter 30 modulates it into an ADPCM signal based on the inputted digital signal. An output from the code converter 30 is inputted to a voice detector 40, and the presence of the conversation voice in a sound signal is decided. The predictive coefficients a1, a2 from the adaptive predictor 35 of the code converter 30 are inputted to frame forming units 41, 42 constituting the voice detector 40. These frame forming units 41, 42 divide the inputted coefficients a1, a2 into short frames respectively, and further, calculate the average value of the predictive coefficients, a1, a2 at every frame by average value calculators 43, 44. Thereafter, the average value of the coefficients a1, a2 is inputted to a sound/silence decision device 45, and the presence of the conversation voice in the sound signal is decided.