Leveraging Transfer Learning for Speech Emotion Recognition Using ResNet50

This paper introduces a speech emotion recognition (SER) system based on transfer learning with ResNet50. It uses the deep hierarchical features of Mel spectrograms of speech signals to precisely classify emotions. We improve feature learning while ensuring computational efficiency by freezing the first 143 layers of ResNet50 and fine-tuning the last layers. The model is trained on the TESS dataset, which contains seven emotions: anger, disgust, fear, happiness, neutral, surprise, and sadness. With data augmentation and optimized hyperparameters, the suggested model obtains a classification accuracy of 98.60\%, surpassing traditional SER methods. The experimental results prove that transfer learning based on ResNet50 can greatly enhance the generalization and robustness of SER models, which makes it an ideal choice for real-world human-computer interaction applications.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-19
DOI
https://doi.org/10.21203/rs.3.rs-11178272/v1
Primary Topic
Emotion and Mood Recognition
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Leveraging Transfer Learning for Speech Emotion Recognition Using ResNet50

P Siva Prasad, Veeranjaneyulu Janagam
Research Square
Emotion and Mood Recognition
preprint

Leveraging Transfer Learning for Speech Emotion Recognition Using ResNet50

P Siva Prasad, Veeranjaneyulu Janagam
preprint en

Abstract

This paper introduces a speech emotion recognition (SER) system based on transfer learning with ResNet50. It uses the deep hierarchical features of Mel spectrograms of speech signals to precisely classify emotions. We improve feature learning while ensuring computational efficiency by freezing the first 143 layers of ResNet50 and fine-tuning the last layers. The model is trained on the TESS dataset, which contains seven emotions: anger, disgust, fear, happiness, neutral, surprise, and sadness. With data augmentation and optimized hyperparameters, the suggested model obtains a classification accuracy of 98.60\%, surpassing traditional SER methods. The experimental results prove that transfer learning based on ResNet50 can greatly enhance the generalization and robustness of SER models, which makes it an ideal choice for real-world human-computer interaction applications.

Research Square
Vignan's Foundation for Science, Technology & Research (IN)
Emotion and Mood Recognition
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon