TANUDJAJA, FRANSISKUS JONATHAN (2023) KLASIFIKASI JENIS EMOSI MELALUI UCAPAN MENGGUNAKAN METODE CONVOLUTIONAL NEURAL NETWORK. Undergraduate thesis, UNIVERSITAS PEMBANGUNAN NASIONAL "VETERAN" JAWA TIMUR.
|
Text (Cover)
19081010062-cover.pdf Download (1MB) | Preview |
|
|
Text (Bab 1)
19081010062-bab1.pdf Download (86kB) | Preview |
|
![[img]](http://repository.upnjatim.ac.id/style/images/fileicons/text.png) |
Text (Bab 2)
19081010062-bab2.pdf Restricted to Repository staff only until 18 July 2025. Download (309kB) |
|
|
Text (Bab 3)
19081010062-bab3.pdf Restricted to Repository staff only until 18 July 2025. Download (535kB) |
|
|
Text (Bab 4)
19081010062-bab4.pdf Restricted to Repository staff only until 18 July 2025. Download (1MB) |
|
|
Text (Bab 5)
19081010062-bab5.pdf Download (11kB) | Preview |
|
|
Text (Daftar pustaka)
19081010062-daftarpustaka.pdf Download (82kB) | Preview |
Abstract
Speech is the preferred and most frequently used method by humans to communicate with each other. Speech itself contains varied information where besides being able to find out the message and intent a person, we can also know the emotional state of that person. Emotional expression in a conversation plays an important role in putting emphasis on the information conveyed becomes more powerful than just words just. Recognizing emotions through speech is very useful and can be applied into various fields such as cognitive science, psychology, neuroscience, call center and other fields. Therefore, the Convolutional Neural Network algorithm will be used (CNN) to classify a person's emotions by their speech. Objective The main objective of this research is to find a CNN model that gives the best performance in classifying emotions into 8 classes, namely, emotions neutral, calm, sad, happy, scared, angry, disgusted and surprised. The CNN model is distinguished based on input data using the feature extraction method different to find the best model. The benchmark for determining the model the best will be done using the value of accuracy, f1-score, precision and recall. The voice dataset used is kaggle, namely the RAVDESS dataset consists of 1440 audio files. From the test results obtained the best model with the ratio of data sharing 80% for training data, 10% for validation data and 10% for test data using the Mel-Frequency Cepstral Coefficients (MFCC) feature extraction method with an average value of 70% accuracy followed by an average value of precision and recall is 68% and 67% respectively. For the most frequently guessed emotion class correctly are the average emotions of anger, surprise, sadness and calm the prediction is correct by 77%.
| Item Type: | Thesis (Undergraduate) | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Contributors: |
|
||||||||||||
| Subjects: | T Technology > T Technology (General) | ||||||||||||
| Divisions: | Faculty of Computer Science > Departemen of Informatics | ||||||||||||
| Depositing User: | Fransiskus Jonathan Tanudjaja | ||||||||||||
| Date Deposited: | 24 Jul 2023 07:05 | ||||||||||||
| Last Modified: | 24 Jul 2023 07:05 | ||||||||||||
| URI: | http://repository.upnjatim.ac.id/id/eprint/15438 |
Actions (login required)
 |
View Item |