378-P: Human Voice Is Modulated by Hypoglycemia and Hyperglycemia in Type 1 Diabetes

Abstract

Anecdotal evidence suggests that high pitched voice may be a clinical symptom of hypoglycemia. We conducted a pioneer study aiming at assessing voice variables during hypoglycemia and hyperglycemia. We recruited 10 non-smoking individuals with type 1 diabetes (4 women, 6 men; mean [±SD] age 40.2±10.4 years, diabetes duration 21.4±9.9 years, HbA1c 8.0±1.8%) who were using real time continuous glucose monitoring system (rt-CGM, Guardian Connect, Medtronic) for mean 5.2±0.6 days in controlled hospital setting. At regular intervals during normoglycemia and additionally at blood glucose <70 mg/dl or >200 mg/dl each patient was instructed to record with a professional voice recorder (LS-14, Olympus) a standardized sentence and the sound of vowel ‘a’ for 3 seconds. The voice samples (n=177) were analyzed against CGMS data with the DiagnoScope software (DiagNova, Wroclaw, Poland). Number of fundamental PERiods (PER), time of fundamental PERiods (PERTime), fundamental frequency (F0), energy (E), amplitude of fundamental frequency (AF0), indicator of voiced/phonation probability (Voiced), simple voice quality (SimpleQ), relative average perturbation (RAP), shimmer (Shimm), amplitude perturbation quotient (APQ), four formant frequencies (F1-F4), harmonic perturbation quotients (HPQ), residual to harmonic ratio (R2H), unharmonic to harmonic ratios (U2H), subharmonic to harmonic ratio (S2H), noise to harmonics ratio (NHR), 1st to 4th harmonic to all energy ratio (Fx1-Fx4) were analyzed. In women during hypoglycemia E, AF0, Voiced, F1, F4, R2H, Fx3, Fx4, while in hyperglycemia - RAP and F2 were significantly altered when compared to normoglycemia. In men in hypoglycemia the differences were found in PER, PERTime, Voiced, SimpleQ, Shimm, APQ, F2, U2H, S2H, Fx2, NHR, while in hyperglycemia - PERTime, F1, HPQ, U2H, Fx2 (all p<0.05). We report for the first time that hypoglycemia and hyperglycemia modulate human voice. This phenomenon may offer potential for early detection and prevention of hypoglycemia. Disclosure L. Czupryniak: None. E. Sielska-Badurek: None. A. Niebisz: None. M. Sobol: None. M. Kmiecik: None. K. Jedra: None. E. Szymanska-Garbacz: None. K. Niemczyk: None. Funding Polish Ministry of Science and Higher Education

Identifying individuals with recent COVID-19 through voice classification using deep learning

“…Currently there is no consensus on which kinds of speech or voice are more suitable for use as voice markers. For example, voice biomarkers for diabetes are varied in the literature and include matched fragments of speech 23 , free speech 24 or vowel sounds 25 . The relative accuracy of using different kinds of human voices for such purposes are currently unclear.…”

Impact of Diabetes Mellitus on Voice: A Methodological Commentary

“…The authors have no competing interests to declare Energy, amplitude of fundamental frequency, indicator of voice/phonation probability, formant frequencies of F1, F2, F4, residual to harmonic ratio, harmonic to all energy ratio, relative average perturbation (RAP), number of fundamental periods, time of fundamental periods, simple voice quality, shimmer, amplitude perturbation quotient (APQ), unharmonic to harmonic ratio, noise to harmonic ratio (NHR) 13 . Static changes due to diabetes reported Fundamental frequency 5, jitter, RAP, shimmer, APQ, smoothed APQ, NHR 4 .…”

“…Recently the results were published regarding the temporal changes in the voices of diabetic patients due to glucose swings 6,7,12,13 , which were revealed via a comparison of voice data in diabetic patients in the conditions of hypo-and hyperglycemia. Table III summarizes the literature with regards to the acoustic parameters that were reported to be affected as a result of glucose swings and the ones that form the descriptive space of the voice changes in diabetes that are listed in Section "Acoustic Analysis".…”

“…Table III summarizes the literature with regards to the acoustic parameters that were reported to be affected as a result of glucose swings and the ones that form the descriptive space of the voice changes in diabetes that are listed in Section "Acoustic Analysis". In order to study the effects of glucose swings in the voices of diabetics, Czupryniak at al 13 applied a statistical analysis to the same gold set of acoustic parameters. (Note that although 13 is an abstract and not a full peer reviewed paper, other studies 6,19 , too, report that glucose swings are non-randomly related to acoustic parameters 6 and, furthermore, that glucose value can be approximated from voice clues on a large corpus of patients 19 .)…”

“…In order to study the effects of glucose swings in the voices of diabetics, Czupryniak at al 13 applied a statistical analysis to the same gold set of acoustic parameters. (Note that although 13 is an abstract and not a full peer reviewed paper, other studies 6,19 , too, report that glucose swings are non-randomly related to acoustic parameters 6 and, furthermore, that glucose value can be approximated from voice clues on a large corpus of patients 19 .) It can be seen that the parameters in the description of static and temporal changes overlap: fundamental frequency, relative average perturbation, shimmer, amplitude perturbation quotient, and noise to harmonic ratio.…”