摘要 |
PROBLEM TO BE SOLVED: To discriminate the voice state in a short time and with high accuracy by setting a prescribed ratio increasing a threshold value based on the difference between the minimum value of the level of an input signal in a prescribed time in the past and the threshold value of a present time to quicken the adaptation of the threshold value. SOLUTION: When an input signal is inputted, a frame energy calculating part 2 calculates energies being levels for every frame. A threshold value calculating part 3 calculates a threshold value based on the output of the frame energy calculating part 2 and the output of the threshold value calculating part 3. Then, a comparator part 4 compares the output of the energy calculating part 2 and the output of the threshold value calculating part 3 and outputs results. Moreover, a non-voice frame counter 5 counts the consecutive number of non-voice frames being frames in which energies of input signal are smaller than the threshold value and when the counted result continues in the number of not less than a prescribed number, the counter 5 discriminates that the input signal is a non-voice and outputs a discriminated result. |