Investigating Deep Learning Architectures for Robust Speech Emotion Recognition: A Study of LSTM, CNN, and CLSTM Models

A key step in affective computing and interaction between humans and computers is speech emotion recognition (SER). Deep learning architectures have demonstrated promising performance in a variety of natural language processing applications, including voice emotion identification. In this paper, we investigate the efficacy of three deep learning architectures are Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), and Convolutional LSTM (CLSTM) for robust voice emotion identification. We run studies on publically available speech emotion datasets, using pre-processing techniques like feature extraction and normalization. To examine the resilience of each model, we train and evaluate its performance across multiple emotion classes and environmental situations. In addition, we look into how hyperparameter tweaking and model architectural changes affect each model's recognition accuracy and computing efficiency. Our findings shed light on the merits and limits of each architecture, as well as recommendations for selecting the best deep learning strategy for robust voice emotion identification applications. Overall, this work advances the technology in speech emotion recognition and enables the creation of more effective human-computer interaction systems capable of effectively interpreting and responding to users' emotional states.

Task

Speech emotion recognition

Data

Publicly available speech emotion datasets

Models

LSTM, CNN, Convolutional LSTM (CLSTM)

Preprocessing

Feature extraction and normalization

Evaluated on

Multiple emotion classes and environmental conditions; accuracy and computing efficiency

Publisher

IEEE

Published

November 23, 2024

Pages

1-6

DOI

10.1109/ICEC59683.2024.10837169

Cited by

3 (Google Scholar, Oct 2026)

P. M. Dasude, U. Anand, A. Nayak, D. Pal, "Investigating Deep Learning Architectures for Robust Speech Emotion Recognition: A Study of LSTM, CNN, and CLSTM Models," in 2024 International Conference on Intelligent Computing and Emerging Communication Technologies (ICEC), IEEE, Nov. 2024, pp. 1–6, doi: 10.1109/ICEC59683.2024.10837169.