Voice authentication can be done because there are physical differences in the voice production organs of each person. The user's spoken sound pattern can be used as a voice command as desired. Some features such as accents, intonation, and the way pronunciation produce different patterns. For identification verification, voice data is divided into two groups: voice with defined text and voice with free text. Sounds resulting from the pronunciation of a particular word can be changed from analog to digital form. This change process will result in representation in vector form. One technique in voice recognition classification is the Support Vector Machine (SVM). The study aims to develop SVM algorithms to create free text-based speech patterns, recognition models. The sound pattern classification process uses three kernels for the data set so that the comparison results will be more accurate. The highest accuracy in the linear kernel is found in the 4th loop in the third fold with an accuracy rate of 94.40%. While in the polynomial kernel the highest accuracy at the 6th iteration of the second fold with an accuracy of 96.80%. The highest accuracy rate is found in the RBF kernel on the 8th loop of the third fold with 98.20% accuracy. These test results prove the RBF kernel has the best level of accuracy in free text-based speech recognition.