SISTEM PENGENAL EMOSI BERBASIS SUARA MENGGUNAKAN EKSTRAKSI CIRI FAST FOURIER TRANSFORM

NUR AIZZUN, SAJIDAH (2022) SISTEM PENGENAL EMOSI BERBASIS SUARA MENGGUNAKAN EKSTRAKSI CIRI FAST FOURIER TRANSFORM. Skripsi thesis, Institut Teknologi Dirgantara Adisutjipto.

	Text (ABSTRAK) 18010006_ABSTRAK.pdf Download (31kB)
	Text (BAB I) 18010006_BAB I.pdf Download (113kB)
	Text (DAFTAR PUSTAKA) 18010006_DAFTAR PUSTAKA.pdf Download (71kB)
	Text (FULL SKRIPSI) 18010006_compressed.pdf Restricted to Repository staff only Download (886kB)

Abstract

Emotion recognition is a process to identify emotions in humans. Where emotions can be recognized through words, voice intonation, facial expressions and body language. Along with the development of science and technology, it has an impact on human life, one of which is the emotion recognition system. In this final project the author designed an emotion recognition system based on the human voice using Matlab software with feature extraction used fast fourier transform. Fast Fourier transform is a simple feature extraction process by converting signals in discrete time domain into frequency time domain. This study focuses on four classes of emotions: angry, neutral, happy, and sad. The first stage starts from the voice data retrieval process followed by preprocessing, which includes normalization, frame blocking, and windowing. Then the values that have been obtained are forwarded to enter the fast Fourier transform and dynamic time warping feature extraction process as the last process in the form of classification to determine the output by comparing the value of the training data and test data to find the minimum value. The minimum value comes from optimal path warping, which is looking for the shortest path from the difference in data comparison. The total voice data used is 32 voice signals with 20 used as training data and 12 as test data where the data is divided equally for each emotional class. Then obtained 5 sound signals as training data and 3 sound signals as test data for each emotion class including angry, neutral, sad, and happy. From the results of the test using 3 test data, which were compared with 20 training data, it was found that words were successfully recognized for angry emotions 2 words, neutral emotions 2 words, sad emotions 3 words, and happy emotions 2 words. So from these results obtained the highest accuracy value on sad emotions of 100% and other accuracy values obtained on average with 66.67%. So from the overall accuracy value obtained for all emotion classes, the average accuracy value is 75%

Actions (login required)

View Item

Downloads