Parkinson's disease (PD) is a worldwide health problem with a wide range of motor symptoms and non-motor symptoms signs caused by the death of dopamine-producing neurons. Voice loss shows up early, which is very helpful for making a quick evaluation. Speech signal processing shows voice as a deep phenotype, which makes it possible for digital measures to speed up evaluations. Traditional testing methods that use neuroimaging and standardized scales have problems, which makes it even more important to come up with new ideas. Using different datasets, machine learning (ML) and deep learning (DL) are revolutionizing how we can predict PD. AI systems improve the accuracy of diagnoses by spotting small patterns. Deep learning, especially convolutional and recurrent neural networks, is very good at catching the complicated voice patterns that come with Parkinson's disease getting worse. There are still problems with using AI to predict PD, such as making sure that data is consistent, models can be understood, and there are social issues to think about. The search for a complete non-invasive clinical screening method goes on, which calls for new tools. This survey reviews recent research findings of artificial intelligence techniques for Parkinson's disease prediction, analyzes the employed methods, techniques, and challenges encountered by researchers, and elucidates their contributions to the field.