Dementia presents a significant challenge due to its multifaceted nature, impacting memory, cognition, and decision-making abilities, thereby impeding daily functioning. Early detection of dementia is crucial for implementing interventions that can potentially slow its progression or alleviate its impact on individuals and their caregivers. However, diagnosing dementia is often complex, resource intensive and requires extensive cognitive assessments and medical evaluations. This study proposes a novel approach to dementia detection by leveraging speech signals and employing deep-learning models. Unlike previous research that relied on a combination of speech and linguistic data, this study focuses exclusively on analysing speech features. By refining the speech characteristics, the study aims to identify a brief set of features that can aid in recognizing dementia. The integration of machine learning (ML) and deep learning (DL) techniques further enhances the potential of this approach. ML algorithms can help in identifying patterns and relationships within the dataset, while DL models, with their ability to process large amounts of data, offer a deeper understanding of complex speech patterns associated with dementia. The dataset used for this research is sourced from the Dementia Bank, which provide samples of speech signals from individuals with varying degrees of cognitive impairment. The outcomes obtained through the proposed method show promising results when compared with existing studies on utilizing speech for dementia recognition. By focusing solely on speech features and employing advanced ML and DL techniques, this study contributes to the growth of research that aims towards improving early detection and management of dementia. This approach unequivocally presents a cost-effective and non-evasive screening method for dementia, ensuring earlier diagnosis and intervention, thus enhancing the daily lives for both affected individuals as well as their families.