An interesting paper that inspired my research ideas on Traditional Chinese Medicine (TCM) auditory diagnosis (listening to voice). The voice naturally differs under various conditions, but explaining it with TCM theory is not easy to understand. The results of this paper are quite good (much better than the model I made 
), and more importantly, it explains to some extent the mechanism behind these results. In the future, explaining TCM will also need to integrate anatomy and pathophysiology.
The variation in these features found that women with T2DM reported a slightly lower pitch with less variation, and men with T2DM reported slightly weaker voices with more variation. These differences likely stem from differences in disease symptom manifestations between the sexes. For example, muscle weakness and atrophy, which have been linked to vocal weakness and instability, 7 occur in T2DM and are more common in men with T2DM than in women with T2DM. 37 On the contrary, women with T2DM are more likely to experience high extracellular water content and edema.38 Swelling and edema of the vocal cords reduce the pitch and vibratory characteristics, resulting in a parameter decrease similar to what was seen in our results. 39 Laryngoscopy and visualization of the vocal cords in T2DM should be performed in future studies to confirm these findings. Furthermore, cognitive function decline and major depressive disorder (MDD) occur at a higher prevalence in women with T2DM than in men with T2DM, 9,40 and peripheral neuropathy occurs at a higher prevalence in men with T2DM. 41 Cognitive impairment has been shown to have a significant effect on the voice with strong predictive capabilities, 10 and MDD has been linked to voice changes such as slower speech and a lower pitch.13 Sex differences in T2DM have become increasingly prominent, as seen in the contrasting predictive features, and future research should carefully account for this for a more comprehensive insight.
These feature variations found that women with T2DM reported a slightly lower pitch and less variation, while men with T2DM reported slightly weaker voices with more variation. These differences may originate from differences in symptom manifestations between sexes. For example, muscle weakness and atrophy associated with vocal weakness and instability occur in T2DM patients and are more common in males with T2DM than in females with T2DM. Conversely, women with T2DM are more prone to high extracellular water content and edema. Swelling and edema of the vocal cords reduce pitch and vibratory characteristics, leading to parameter reductions similar to those observed in our results. Laryngoscopy and vocal cord visualization should be conducted in future studies to confirm these findings. Additionally, cognitive decline and major depressive disorder (MDD) have a higher prevalence in women with T2DM than men with T2DM, while peripheral neuropathy has a higher prevalence in men with T2DM. Cognitive impairment significantly affects voice with strong predictive ability, and MDD is linked to voice changes such as slower speech and lower pitch. Sex differences in T2DM are becoming more pronounced, as reflected in contrasting predictive features, and future research should carefully consider this for more comprehensive insights.
Original Article Link
https://www.mcpdigitalhealth.org/article/S2949-7612(23)00073-1/fulltext
Introduction from Zhihu
https://zhuanlan.zhihu.com/p/665169449
Recently, Canadian medical researchers trained a machine learning artificial intelligence to identify 14 voice differences between patients with type 2 diabetes (T2DM) and non-diabetic patients to diagnose T2DM.
The auditory features focused on by the AI include subtle changes in pitch and intensity that the human ear cannot distinguish. These are then paired with basic health data collected by researchers, such as age, sex, height, and weight.
Jaycee Kaufman, the first author of the paper and research scientist at Klick Labs, the company planning to commercialize this software, explained: “Our study highlights significant voice differences between patients with T2DM and those without.”
Regarding the company’s AI, Kaufman hopes it will “change the way the medical field screens for diabetes.”
Common current tests for diagnosing T2DM include glycated hemoglobin (A1C) testing, fasting blood glucose (FBG) testing, and oral glucose tolerance testing (OGTT), all of which require onsite testing.
Kaufman added: “Current testing methods may require significant time, labor, and economic costs. Voice technology has the potential to completely eliminate these barriers.”
Scientists from Klick Applied Science in Canada collaborated with faculty from Ontario Tech University to train AI using 267 voice recordings from Indian participants.
About 72% of participants (79 females and 113 males) were diagnosed as non-diabetic. The other participants (18 females and 57 males) were diagnosed with type 2 diabetes.
All participants recorded a short phrase six times a day for two weeks, resulting in a total of 18,000 recordings. Scientists then identified 14 voice differences between patients with T2DM and non-diabetics.
Four of these differences helped the AI diagnose T2DM more accurately. Trial data showed the AI was better at diagnosing women accurately: 89% of women and 86% of men were correctly diagnosed as having T2DM.
The study found that “pitch” and “pitch standard deviation” are useful features for diagnosing all participants. However, “relative average perturbation (RAP) jitter” is more useful for females. “Intensity” and the “11-point amplitude perturbation quotient (APQ11) shimmer” are very useful for diagnosing males.
The study pointed out: “For females, predictive features are average pitch, pitch SD, and RAP jitter, while for males, average intensity and APQ11 shimmer are used. Simply put, these feature variations found women with T2DM reported slightly lower pitch and less variation, whereas men with T2DM reported slightly weaker voice and more variation. These differences may stem from differences in disease symptom manifestations between sexes.”
Kaufman commented that these differences between male and female voices, discovered through AI signal processing, are “surprising.”
Researchers concluded: “Voice analysis shows potential as a T2DM prescreening or monitoring tool, especially when combined with other disease-related risk factors.”
The study was published in the journal Mayo Clinic Proceedings: Digital Health.
Reference source: https://www.diabetes.co.uk/news/2023/oct/say-what-ai-can-diagnose-type-2-diabetes-in-10-seconds-from-your-voice.html