2020³â ÄÄÇ»ÅÍÁ¾ÇÕÇмú´ëȸ
Current Result Document :
ÇѱÛÁ¦¸ñ(Korean Title) |
A Hybrid approach for speech emotion recognition using 1D-CNN LSTM |
¿µ¹®Á¦¸ñ(English Title) |
A Hybrid approach for speech emotion recognition using 1D-CNN LSTM |
ÀúÀÚ(Author) |
Sanghoon Lee
Taeho choi
Minjae Joo
Sunkyu Kim
Inggeol Lee
Junseok Choi
ÀÌ»óÈÆ
ÃÖÅÂÈ£
ÁÖ¹ÎÀç
±è¼±±Ô
ÀÌÀ×°É
ÃÖÁؼ®
°Àç¿ì
Gayrat Tangriberganov
Tosin A. Adesuyi
Byeong Ma
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 47 NO. 01 PP. 0833 ~ 0835 (2020. 07) |
Çѱ۳»¿ë (Korean Abstract) |
|
¿µ¹®³»¿ë (English Abstract) |
Speech is an important aspect of human interaction in which relationship and emotions can be expressed. The ability to learn and recognize human emotions through speech has become an area of interest in the field of human-machine interaction and machine learning. Through the use of private and publicly available dataset, several researches on speech emotion recognition (SER) has been successfully carried out. However, their recognition results requires improvement, because important speech emotion features are not well captured during training. Hence, we proposed a 1D-CNN LSTM for speech emotion recognition to assist in the capturing of global and local features. The 1D-CNN part is responsible for learning salient features required for recognition while the LSTM is delegated for classification based on the extracted features received from the 1D-CNN. We experimented our hybrid approach with the raw audio from Berlin EmoDB dataset and an average accuracy of 95.5% and validation accuracy of 63.7 % were achieved. Our result an improvement over the existing studies. |
Å°¿öµå(Keyword) |
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|