Voice based gender recognition using deep learning

Voice based gender recognition using deep learning

Predicting age through audio is essential for tailored user experiences, such as content recommendations and targeted advertisements. Better accessibility features for those with age-or gender-specific needs are also made possible by it. This paper presents a comprehensive approach to gender classif...

Full description

Saved in:
Bibliographic Details
Published in:2024 15th International Conference on Computing Communication and Networking Technologies (ICCCNT) pp. 1 - 8
Main Authors: Neelamegan, M, Kanimozhi Selvi, C S, Shanmugam, Manjupargavi, Sabari, S, Vignesh, T
Format: Conference Proceeding
Language:English
Published: IEEE 24-06-2024
Subjects:
Accuracy
Biometrics
Convolutional Neural Network (CNN)
Convolutional neural networks
Data mining
Data models
Deep learning
Early stopping
Feature extraction
Librosa
Long short term memory
LSTM Model
Mel-Frequency Cepstral Coefficient
Neural networks
Principal component analysis
ResNet
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Predicting age through audio is essential for tailored user experiences, such as content recommendations and targeted advertisements. Better accessibility features for those with age-or gender-specific needs are also made possible by it. This paper presents a comprehensive approach to gender classification using voice data, employing deep learning techniques such as Long Short Term Memory (LSTM), Convolutional Neyral Network (CNN), and Residual Network (ResNet) models. The work begins with the preprocessing of audio data, extracting Mel-Frequency Cepstral Coefficients (MFCCs) as features. To address class imbalance, Random Oversampling and Random Undersampling techniques are applied. Then, using the resampled data, LSTM, CNN, and ResNet models are built and trained. Model evaluation includes metrics such as classification reports, confusion matrices, and (Receiver Operating Characteristic Curve) ROC curves. The outcomes show how successful the suggested models are, with considerations for oversampling and undersampling strategies. The results, based on training the model on the SLR45 dataset, demonstrate an impressive accuracy of approximately 99% for gender classification by CNN.
ISSN:2473-7674
DOI:10.1109/ICCCNT61001.2024.10725158