Voice recognition using machine learning techniques: A literature review

This study aims to compare different machine learning methods and determine the most effective method by reviewing the literature on machine learning methods used in the field of voice recognition. Within the scope of the study, 30 studies were analyzed in detail by reviewing the current literature published between January 2023 and March 2025. The study concluded that deep learning-based approaches, especially Transformer architectures and end-to-end learning models, exhibit higher accuracy and robustness compared to traditional methods. However, instead of a single "best" approach for an ideal voice recognition system, a combination of different approaches may be more appropriate depending on the application scenario, available resources and target user group. Despite the significant progress made in the development of voice recognition systems, several problems and limitations still exist, such as the need for large amounts of labeled data and performance degradation in noisy environments. For future research, semi-supervised and self-supervised learning approaches for low-resource languages, hybrid model architectures, multi-task and multi-modal learning approaches, neuromorphological computational approaches, and standardized evaluation metrics are proposed.