Fusion of MFCC Features and CNN Architectures for Multi-Dataset Speech Emotion Recognition

View at publisher Summarise with VIRAT
Authors: Narinder Kaur, Neha Dhiman
Cited by: 0
References: 18

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Emotion identification from speech is essential for enhancing human-computer interaction, allowing systems to react with contextual empathy. This article presents a convolutional neural network (CNN)-based system that uses Mel-frequency cepstral coefficients (MFCCs) as input characteristics to categorise speech patterns according to emotions. Our approach employs the spatial structure of MFCCs to inherently discern patterns through deep convolutional layers, rather than relying on handcrafted features and basic classifications. We developed and enhanced a CNN architecture, subsequently utilising typical emotional speech datasets to evaluate its efficacy. Our experimental data indicate that our model significantly outperforms standard classification approaches. This demonstrates that deep learning is an effective method for detecting nuanced emotional signals in speech.

Fields of Study & Subjects

Emotion and Mood Recognition Speech Recognition and Synthesis Speech and Audio Processing

Publication details

Journal / source
International Conference on Signal Processing and Electronics Design (ICSPED)
Pages
71–76
Publication date
23 January 2026
Language
EN