Segmenting Speech Signals May Effectively Detect Presence of Common Cold, Flu

Monitoring unvoiced speech segments could help diagnose common cold and associated disorders, according to a study published in the National Library of Medicine. Understanding speech in the context of common cold and flu could facilitate remote monitoring of a patients’ health.

“The analysis and classification of cold speech may be useful in the diagnosis of common cold and other related illnesses to stop the spread of these viral infections,” the study authors wrote.

A speech signal contains linguistic (communicated) information and paralinguistic (emotion, health state, age, gender) information that are both involved in physiological and cognitive systems. Speech signals and vocals may even offer insight into the mental and physical conditions of the body, according to the investigators.

“Speech is produced as a consequence of the vocal tract’s linear filtering of stimulation source data. Because the vocal tract is engaged during speech production, the acoustic properties of cold speech vary from those of normal speech,” the study authors wrote.

Cold speech is pathological because it affects the nasal and esophagus—it reflects cold or flu symptoms such as stuffy nose, hoarse voice, coughing and sneezing—and can alter speech signal. Researchers evaluated regions of voiced (vibrating vocal cords) and unvoiced (non-vibrating vocal cords) speech to classify cold versus healthy speech, while also determining a framework for identifying cold speech.

Identifying 630 participants from the Upper Respiratory Tract Infection Corpus (URTIC) database out of Germany, researchers listened to a voice recording from everyone. Voice recordings consisted of a binary 1-item measure based on the Wisconsin Upper Respiratory Symptom Survey (WURSS-24) and evaluated on a scale of 0 (not sick) to 7 (very ill).

Researchers collected complete active speech (CAS)—non-silent speech segments comprised of voiced and unvoiced speech regions. They collected frame-wise mel frequency cepstral coefficients ([MFCC] from CAS, unvoiced, and voiced speech) to then segment speech, and also used a support vector machine (SVM) to evaluate the voiced and unvoiced regions of speech.

The results show that unvoiced regions of speech signal have less frames than CAS or voiced regions of speech, suggesting that it can be an effective and efficient tool to detect the presence of upper respiratory tract infections (URTIs), such as the common cold and flu. URTIs impact millions of individuals every year and may severely affect up to 5 million people annually.

“If someone is merely interested in common cold detection with the minimum effort and complexity, they can evaluate only the unvoiced portion of the speech,” the study authors wrote.

Further, extracting voiced and unvoiced segments of speech to determine the presence of a cold can be used in smart devices to monitor cold and related conditions to stop the spread of illness, according to the study.

“The results show that the feature extracted from voiced and unvoiced segments shows the same discrimination capability for cold and healthy speech,” the researchers wrote in the report.

Reference

Warule, Pankaj, Prasad Mishra, Siba, Deb, Suman. Significance of voiced and unvoiced speech segments for the detection of common cold. Natl Lib of Med. Nov 15, 2022. Accessed on Dec 16, 2022. doi: 10.1007/s11760-022-02389-8