Comparison Euclidean Distance and Manhattan Distance as Classification in Speech Recognition System
Muhammad Ryandy Ghonim Asgar, Risanuri Hidayat, Agus Bejo · 2023
One of the uses of a digital system is a speech recognition system.Feature extraction and classification is important step in speech recognition system process.Mel Frequency Cepstrum Coefficient (MFCC) feature extraction is a popular feature extraction used in speech recognition system, while one of the most popular classification technique is K Nearest Neighbour (KNN).There are many KNN classification techniques, but the most commonly used are the Euclidean Distance and Manhattan Distance.Research on speech recognition system in Indonesia and in particular the Indonesian speech recognition system is still very limited, far from the recognition system in English.Therefore, this paper proposes a comparison of the best accuracy generated by the classification between Euclidean distance and Manhattan distance using MFCC as a feature extraction in Indonesian speech recognition system.The model and testing of the proposed system used is 120 data, with 0 to 9 voice signals in Indonesian.By using the 13 coefficients from the MFCC and using 5-fold cross validation to achieve generalized results, the Euclidean distance is able to outperform the accuracy obtained by the Manhattan distance by a value of 88%.