Dementia is a serious public health concern. It does not have any cure. Early detection of dementia is, thus, critical for effective symptom management as well as delaying cognitive and functional decline. This paper focuses on detecting onset of dementia using text and speech features provided by two publicly available datasets from the AAAI 2022 hackallenge. Our approach resulted in developing ACOUSTICS (AutomatiC classificatiOn of sUbjectS with demenTIa and healthy Controls using text transcriptions and Speech data)—an ensemble model with two deep learning-based architectures for text and speech analysis. ACOUSTICS achieved 89.8% accuracy when classifying individuals with dementia and health controls. Our approach outperforms current state-of-the-art methods in dementia detection.