On the effect of feature compression on speech emotion recognition across multiple languages

• Main Authors: Alghifari, Muhammad Fahreza, Gunawan, Teddy Surya, Nik Hashim, Nik Nur Wahidah, Wan Nordin, Mimi Aminah, Kartiwi, Mira
• Format: Book Chapter
• Language: English; English
• Published: Springer 2020
• Subjects: TK7885 Computer engineering
• Online Access: http://irep.iium.edu.my/84767/2/Paper_185.pdf; http://irep.iium.edu.my/84767/1/Acceptance%20Letter_DrTeddy_IIUM.pdf; http://irep.iium.edu.my/84767/; https://im3f2020.ump.edu.my/index.php/en/
• Institution: Universiti Islam Antarabangsa Malaysia

The ability of computers to recognize emotions from the speech is commonly termed as speech emotion recognition (SER). While in recent years, many studies have been performed, the golden standard has yet to be achieved due to many pa-rameters to consider. In this study, we investigate the effect of speech feature compression of Mel-frequency Cepstral Coefficient (MFCC) across four lan-guages – English, French, German, and Italian. The classification was performed using a deep feedforward network. The proposed methodology has shown to have significant results when tested using a network which was trained in the same language, and up to an accuracy rate of 80.8% when trained using all four languages.