LSTM- and GRU-Based Common Cold Detection from Speech Signal

This paper proposes a neural-network-based detection of common cold disease through speech signal processing. Speech is formed by the movements of vocal cords present in the vocal tract. The vocal tract includes organs like the nasal cavity, the oral cavity, the upper throat (pharynx), and larynx. A common cold is a viral infection that affects the upper respiratory tract, which includes nose, throat, sinuses, and airways. Therefore, a common cold alters a person’s vocal ability. The goal of the proposed study is to create two classification models based on long short-term memory (LSTM) and gated recurrent unit (GRU) that can properly determine from an individual’s voice whether they are infected with common cold. Through the use of these architectures we highlight the role of speech signal processing in the early detection and diagnosis of the common cold. The results show satisfactory accuracy in detecting cold speech. The results are preliminary, and by balancing the dataset and focussing on a wider sample of sick individuals, it may be possible to improve accuracy.