Deep Learning-Based Classification of Spoken English Digits.

Area of Science:

• Speech recognition
• Machine learning
• Artificial intelligence

Background:

• Classifying isolated spoken digits is a fundamental challenge in speech recognition systems.
• Limited research exists on spoken English digit datasets compared to broader spoken language studies.

Purpose of the Study:

• To investigate and compare machine learning algorithms for classifying spoken English digits.
• To propose an intelligent system using deep feedforward neural networks (DFNN) for enhanced spoken digit classification.

Main Methods:

• Employed deep feedforward neural networks (DFNN) with hyperparameter optimization.
• Utilized ensemble methods like Random Forest (RF) and regression methods like Gradient Boosting (GB) for comparison.
• Evaluated algorithms on a public benchmark dataset for spoken English digits.

Main Results:

• The proposed DFNN classifier achieved a validation accuracy of 99.65%.
• DFNN significantly outperformed Random Forest and Gradient Boosting classifiers.
• The developed model demonstrates superior performance compared to existing systems using traditional classifiers.

Conclusions:

• Deep feedforward neural networks offer a highly effective approach for spoken English digit classification.
• The proposed DFNN-based system provides a state-of-the-art solution for this specific speech recognition task.
• Further research can explore advanced deep learning architectures for improved accuracy and robustness.