Author:
Pavithra P, Priya N, Naveenkumar E
Published in
Journal of Science Technology and Research
( Volume , Issue )
Page No: 65-77
Volume , Issue
Article Type:
Published Date:
Published by:
Abstract
In modern days, person-computer communication systems have gradually penetrated our lives. One of the crucial technologies in person-computer communication systems, Speech Emotion Recognition (SER) technology, permits machines to correctly recognize emotions and greater understand users' intent and human-computer interlinkage. The main objective of the SER is to improve the human-machine interface. It is also used to observe a person's psychological condition by lie detectors. Automatic Speech Emotion Recognition(SER) is vital in the person-computer interface, but SER has challenges for accurate recognition. In this work to resolve the above problem, automatic Speech enhancement shows that deep learning techniques effectively eliminate background noise. Using Deep leaning models for four states were created: happy, sad, angry, and intoxicated. Recurrent Neural Network (RNN) algorithm used to reduce the possibility of over fitting by randomly omitting neurons in the hidden layers. The proposed RNN method could be implemented in personal assistant systems to give better and more appropriate state-based interactions between humans. In the simulation results shows Improving accuracy, Time complexity, Error rate is also reduced to using the proposed method.
Keywords
Speech Emotion Recognition (SER), Speech emotion detection, deep leaning, Recurrent Neural Network (RNN), preprocessing, feature extraction.
References
Data not available

ABSTRACT:

Speech Emotion Recognition (SER) is a key component in modern human-computer interaction systems. It enables machines to understand human emotions through speech, improving the quality of communication and user experience. This project focuses on developing an automatic SER model using deep learning, specifically Recurrent Neural Network Based Speech. RNNs are effective for processing sequential data and reducing overfitting by randomly omitting neurons in hidden layers. The proposed system is trained to detect four emotional states: happy, sad, angry, and intoxicated. It integrates speech enhancement to eliminate background noise, improving recognition accuracy. The goal is to build a system that offers improved emotion-based responses, suitable for personal assistant applications. Simulation results show increased accuracy, reduced error rate, and lower time complexity when using the RNN model. Recurrent Neural Network Based Speech solution can be applied in lie detectors, user mood analysis, and emotion-aware interfaces, offering more appropriate and intelligent interactions between humans and smart systems.

Recurrent Neural Network Based Speech:

Speech is a natural medium for human communication, and emotions make it expressive and meaningful. Recognizing emotional states in speech is vital for intelligent systems that aim to respond empathetically to users. This study explores Speech Emotion Recognition (SER) using Recurrent Neural Networks (RNN), a deep learning method effective for sequential data. Emotions like happiness, sadness, anger, and intoxication significantly influence pitch, tone, and frequency in speech. Traditional systems lack accuracy and struggle with real-world noise. To address these issues, our approach incorporates automatic speech enhancement, improving emotional clarity. The RNN-based model captures low-level audio features without needing manual tuning. It also supports better generalization by omitting certain neurons during training, avoiding overfitting. SER plays a critical role in personal assistants, lie detection, and psychological analysis. By understanding emotional cues, machines can interact more naturally, providing timely feedback and improving the overall experience. This makes SER an essential part of modern intelligent systems.

Recurrent Neural Network Based Speech emotion detection using Deep Learning

DOWNLOAD

INDEXED IN