Mild Cognitive Impairment (MCI), an early stage of dementia including Alzheimer’s disease, is often marked by memory deficits, speech impairments, and reduced reasoning ability. Speech analysis offers a promising, non-invasive tool for early detection of cognitive decline. However, existing approaches typically rely on pre-trained models, language-specific models, or transcription-based features, limiting their applicability in multilingual and low-resource settings. In this study, we introduce a novel, low-cost framework that eliminates the need for transcriptions or pre-trained language models by leveraging only 14 carefully selected, language-independent acoustic features. These features capture prosodic, phonatory, and temporal speech characteristics known to correlate with cognitive decline and are consistent across both English and Chinese speech. The proposed model trained using traditional machine learning algorithms is highly efficient, requiring approximately 14 s for training, making it well-suited for scalable deployment. Evaluated on the bilingual TAUKADIAL dataset, our framework achieves an unweighted average recall (UAR) of 68.36% for MCI classification and a root mean squared error (RMSE) of 2.59 for MMSE score prediction. The results highlight the framework’s cross-linguistic generalizability, computational efficiency, and clinical potential for accessible cognitive assessment.