In this paper, we investigated the parameter extraction to determine whether to drink or not by comparing voice characteristics before and after drinking using voice analysis. The proposed method extracts the valid frame position before and after drinking and identifies the alcohol by using the deviation. Drinking leads to various changes in the voice signal. Using this feature, the effective frame is a method using the fact that the duration of vocalization is longer than that before drinking and the vocalization rate is relatively high. In order to improve the judgment rate by considering the features of each speaker instead of using only the numerical comparison feature of the effective frame deviation as an advanced method, it is tried to determine before and after drinking by using the distribution. The proposed method is robust to judge whether or not to drink alcoholic beverages even in distorted voice waveforms in various environments as well as existing voice signals. In other words, even if the voice signal is distorted, features such as the rhythm and vocalization state inherent to the characteristic of the waveform remain unlike the waveform characteristic, and the higher the number of data sets, the higher the judgment result of drinking can be obtained. When the speech signal was applied before and after drinking with the speech signal using this method, the judgment rate was robust even in various environments