Access the full text.
Sign up today, get DeepDyve free for 14 days.
Aline Pinto, A. Crespo, L. Mourão (2014)Influence of smoking isolated and associated to multifactorial aspects in vocal acoustic parameters
Brazilian Journal of Otorhinolaryngology, 80
Hansa Banjara, V. Mungutwar, Digvijay Singh, Anuj Gupta (2011)Objective and Subjective Evaluation of Larynx in Smokers and Nonsmokers: A Comparative Study
Indian Journal of Otolaryngology and Head & Neck Surgery, 66
R. Fujiki, Abigail Chapleau, Anusha Sundarrajan, Victoria McKenna, M. Sivasankar (2017)The Interaction of Surface Hydration and Vocal Loading on Voice Measures.
Journal of voice : official journal of the Voice Foundation, 31 2
Julio González, A. Carpi (2004)Early effects of smoking on the voice: a multidimensional study.
Medical science monitor : international medical journal of experimental and clinical research, 10 12
João Teixeira, P. Fernandes (2014)Jitter, Shimmer and HNR Classification within Gender, Tones and Vowels in Healthy Voices
Procedia Technology, 16
D. Pinar, H. Cincik, E. Erkul, A. Gungor (2016)Investigating the Effects of Smoking on Young Adult Male Voice by Using Multidimensional Methods.
Journal of voice : official journal of the Voice Foundation, 30 6
I Guimarães (2005)185
Logoped. Phoniatr. Vocol., 30
B. Tuhanioğlu, S. Erkan, T. Özdaş, Çağrı Derici, Kemal Tüzün, Özgül Şenkal (2019)The Effect of Electronic Cigarettes on Voice Quality.
Journal of voice : official journal of the Voice Foundation
I. Guimarães, E. Abberton (2005)Health and voice quality in smokers: An exploratory investigation
Logopedics Phoniatrics Vocology, 30
G. Fagherazzi, A. Fischer, Muhannad Ismael, V. Despotović (2021)Voice for Health: The Use of Vocal Biomarkers from Research to Clinical Practice
Digital Biomarkers, 5
Y Deng (2020)2545
Stephen Jannetts, F. Schaeffler, J. Beck, Steve Cowen (2019)Assessing voice health using smartphones: bias and random error of acoustic voice parameters captured by different smartphone types.
International journal of language & communication disorders, 54 2
J. Teixeira, C. Oliveíra, C. Lopes (2013)CENTERIS 2013-Conference on ENTERprise Information Systems / HCIST 2013-International Conference on Health and Social Care Information Systems and Technologies Vocal Acoustic Analysis-Jitter , Shimmer and HNR Parameters
(2019)Epic Enterprises LLC. TapeACall: Call Recorder (2019) 20. Cube Apps Limited
H Banjara (2014)99
Indian J. Otolaryngol. Head Neck Surg., 66
W. Decoster, F. Debruyne (2000)Longitudinal voice changes: facts and interpretation.
Journal of voice : official journal of the Voice Foundation, 14 2
Yujiao Deng, Meng Wang, Linghui Zhou, Yi Zheng, Na Li, Tian Tian, Zhen Zhai, Si Yang, Qian Hao, Ying Wu, Dingli Song, Dai Zhang, Jun Lyu, Z. Dai (2020)Global burden of larynx cancer, 1990-2017: estimates from the global burden of disease 2017 study
Aging (Albany NY), 12
R. Young, R. Hopkins (2012)Increasing smokers' risk perception improves CT screening participation
P. Boersma, D. Weenink (2003)Praat: doing phonetics by computer
W Decoster (2000)184
J. Voice, 14
Olivia Murton, R. Hillman, D. Mehta (2020)Cepstral Peak Prominence Values for Clinical Voice Evaluation
American Journal of Speech-Language Pathology, 29
Ouissam Zealouk, H. Satori, Mohamed Hamidi, N. Laaidi, K. Satori (2018)Vocal parameters analysis of smoker using Amazigh language
International Journal of Speech Technology, 21
H. Byeon, Seulki Cha (2020)Evaluating the effects of smoking on the voice and subjective voice problems using a meta-analysis approach
Scientific Reports, 10
JP Teixeira, C Oliveira, C Lopes (2013)Vocal acoustic analysis?Jitter, Shimmer and HNR parameters
Procedia Technol., 9
(2010)Te ara tika guidelines for Māori research ethics: a framework for researchers and ethics committee members
Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations
This study aimed to determine if self-complete at-home recordings could produce audio samples of sufﬁcient quality for use in voice analysis software, and if audio samples of similar or sufﬁcient quality could be extracted from audio-recorded naturalistic phone interviews. Data were obtained from 31 adults aged 18 years and over who smoked. The /a/ sound segment was manually isolated, and analysis functions were used to produce the following values: fundamental frequency, jitter, shimmer, noise ratio, formant 3, and formant 4. The /a/ sound segment was then manually isolated from audio recordings of naturalistic interviews previously conducted by phone. These were analysed in the same way and compared for quality against Evistr-recorded audio samples from the same participants. A third audio sample consisted of an Evistr or phone-recorded sustained phonation of the /a/ sound. Means and standard deviations were calculated for the target vocal parameters. Statistical comparisons for quality of sound segment were conducted for readings, interviews, and vowel phonation and for sound signals extracted via both recording methods. Self-recording by adults who smoked provided audio samples of sufﬁcient quality for analysis of vocal features that have been associated with a clinical outcome. The values obtained for sustained phonation audio samples displayed the least perturbation and noise for the vocal parameters surveyed. Sound signals recorded with smartphones appeared to be affected by electronic interference but have potential for use in diagnostic tools for measuring vocal parameters. Keywords Voice recording · Acoustic analysis · Vocal biomarkers · Larynx · Smoking 1 Introduction of the vocal folds . Smoking has a signiﬁcant effect on laryngeal structures, leading to numerous health issues that 1.1 Voice Contributes to General Health include changes in the laryngeal area or more serious issues such as chronic inﬂammatory changes, which can lead to Speaking fundamental frequency (SFF) decreases as peo- cancer and death . Smoking can cause sinusitis, gastroe- ple age. Speaking frequencies are also affected by particular sophageal disease, and respiratory disease, which includes jobs. For example, a history of working in broadcasting has emphysema, bronchitis, and cancer . The level of deterio- been associated with varying degrees of voice deterioration. ration of the laryngeal and voice is partly determined by how While there are still many questions about the relative impact long a person has smoked, the quantity of cigarettes smoked, of voice and comorbid conditions on general health, it is and alcohol consumption. For example, the risk of devel- known that people who smoke tobacco are at a higher risk of oping laryngeal cancer is greater among people who smoke voice deterioration. Smoking irritates the lining of the larynx, more than 35 cigarettes a day . which can cause dehydration that affects the free vibration 1.2 Larynx Cancer B Marewa Glover email@example.com The larynx, which is part of the throat situated at the entrance 1 of the windpipe (trachea), plays an important role in breath- Centre of Research Excellence: Indigenous Sovereignty and Smoking, PO Box 89186, Torbay, Auckland 0742, New ing and speech. Larynx cancer is one of the most common Zealand cancers of the head and neck. Larynx cancer accounts for 1% Independent Researcher, Auckland, New Zealand of all new cancer cases and deaths globally. Between 1990 123 Acoustics Australia and 2017, the global cases of larynx cancer increased from 1.5 Vocal Biomarkers of Smoking-Related 132,740 to 210,610 . In New Zealand (NZ), in 2018 lar- Deterioration ynx cancer accounted for 1 per 100,000 population of new cancer cases, a slight decline from 1.2 per 100,000 in 2013 The negative impact of smoking on voice has been demon- . While tobacco smoking alone does not cause all larynx strated in two cross-sectional studies [7, 8] that compared cancers, it is estimated that smoking increases the likelihood vocal parameters in adults who smoked, and adults who did of developing larynx cancer. not smoke. Both studies analysed vowel sounds—[a] in Pinto et al. , and [æ], [i], and [*] in Zealouk et al. —in the speech of all participants. Both studies measured mean values for pitch (i.e. fundamental frequency F0), jitter, and shimmer, 1.3 Increased Need for Remote Delivery for the two groups of adults. Additionally, Zealouk et al. of Healthcare included formants values in their comparison, while Pinto et al. considered noise-to-harmonic ratio (NHR). Both stud- The widespread use of lockdown strategies, in response ies observed that the values for fundamental frequency were to the COVID-19 pandemic, has challenged the domi- lower for adults who smoke, whereas shimmer and jitter val- nant face-to-face delivery mode of most healthcare services ues are higher for the group of participants who smoke. It and research, including clinical trials. Lockdown measures was also found that the presence of smoking can lower F3 included shelter-in-place orders for non-essential workers, and F4 values of formants and increase NHR values. that is, that they stay home and limit excursions beyond their Another cross-sectional study on adults who smoked ver- property. sus adults who had never smoked examined the early effects Remote health delivery presents a particular challenge of smoking on the voice quality of 134 young women and for smoking cessation services and studies that use biofeed- men, aged 20–29 years (mean age 22 years) . Vocal back tools, such as practitioner-administered exhaled carbon parameters were measured for the fundamental frequency, monoxide tests to motivate smoking cessation and val- jitter, shimmer, and noise-to-harmonic ratio for the /a/ sound. idate smoking status and health improvement following The values measured for the fundamental frequency F0, abstinence. Some diagnostic sampling tools can be self- frequency perturbation parameters (jitter absolute ‘jitta’), administered at home, such as cotinine in saliva tests, clipping and period-to-period perturbation quotients were noticeably a sample of hair, or collecting a urine sample. Postage and lab affected by cigarette smoking, even among people who had analysis costs can reduce accessibility to these methods. A smoked for 10 years or less. Fundamental frequency was further challenge is that most tests do not distinguish between signiﬁcantly lower in young women who smoked, while fre- nicotine use and tobacco smoking. Sending bodily samples to quency perturbation was signiﬁcantly higher in young men foreign laboratories for genetic analysis is an additional bar- who smoked. However, early stage of smoking was not found rier to use of these methods. For example, some religious and to have detrimental effect on the amplitude of sound sig- cultural groups, including Indigenous peoples, have bioeth- nals (shimmer parameters) or to add noise-to-sound signals ical concerns about the use of their genetic information . (NHR). Therefore, we expect that we will similarly ﬁnd that F0 and jitter parameters will be more sensitive to the effects of smoking than shimmer and NHR parameters, and that the 1.4 Vocal Biomarkers effect of smoking on the F0 and jitter vocal parameters will differ by sex. Voice represents a potentially low-cost, easily accessed, and non-invasive source for obtaining individual biomarkers to aid health diagnoses. The application of voice diagnosis tech- nologies will have more uses and greater value globally if audio samples of sufﬁcient quality can be collected remotely 2 Materials and Methods using a telephone or other digital platforms, and from natural- istic speech. Studies determining the acceptability, potential 2.1 Design efﬁcacy, and identifying parameters of relevance to different conditions in real-world settings are needed to advance this This cross-sectional observation study aimed to determine if: ﬁeld. This paper reports on a cross-sectional study conducted to inform the development and validation of a voice recording 1. Self-complete-at-home recordings produce audio sam- and analysis protocol for potential use in the future smoking ples of sufﬁcient quality for use in voice analysis software cessation studies. Praat  and 123 Acoustics Australia 2. Audio samples of similar or sufﬁcient quality for use in 2.5 Script Praat can be extracted from audio-recorded naturalistic phone interviews. Participants were asked to read the Rainbow Passage.This script was chosen for its sound repetition, unusual consonants A third objective was to identify optimum voice param- and vowel combinations, and having short and long passages eters to inform the design of research on the intervention that help to test speech patterns . potential of voice deterioration information for enhancing risk perception and motivation to stop smoking. 2.6 Data Collection Procedure 2.2 Participants Demographic information (age, sex) and history of smoking data (age began regular smoking, total estimated time off Participants were 49 NZ adults aged 18 years and over who smoking) were extracted from the Voices of the 5% Study smoked and had no immediate intent to stop smoking. They baseline interview data. were drawn from a larger four-year longitudinal qualitative Participants were informed that a recorder was going to study aimed at identifying barriers to stop smoking. be sent to them and they were given an overview of the instructions for its use. The recorder, task instructions, and 2.3 Recruitment (into the Voices of the 5% Study) return-addressed courier bag were then sent to each partici- pant. Advertisements for Voices of the 5% Study participants were Participants’ baseline interviews generally took place placed in NZ print and online media, including social media. before their initial recording of reading the Rainbow Passage. Similarly, notices were distributed to the researchers’ net- Recordings of the vowel sustained phonation took place at works. A snowball method was also used, whereby enrolled a later date—for most participants at the end of a follow-up participants were asked to send the study advertisement or reading. The chronology of data collection is presented in notice to people they knew who might be eligible. Fig. 1. Potentially eligible respondents were sent a participant information sheet (PIS) and consent form by email or post. 2.6.1 Recorded Reading The PIS explained the purpose of the study, what participants would be asked to do, who was conducting the research, and The instruction sheet covered how to use the recorder, to ensure there was no background noise, to sit in a relaxed who was funding the study. The PIS advised the participants that they would be asked to record their voice on a portable position with their chin up, and to say their ﬁrst name, the handheld audio recorder. The PIS stated that: all information time, and what room of their house they were sitting in. When collected from participants would be treated in the strictest prepared as instructed, the participants recorded themselves conﬁdence; no individual would be able to be identiﬁed at reading the Rainbow Passage at a moderate pace, twice. any stage in the publication or presentation of the study’s Upon their return, the audio recordings were then down- ﬁndings; all participant information sheets would be kept in loaded and ﬁled as.wav ﬁles, bit rate 1536 kbps. a locked cabinet separate from de-identiﬁed collected data; and that, following publication of the results of the study, 2.6.2 Phone Interview the raw data would be made available online. In addition, participants were informed that they could withdraw from The second data source for this study was naturalistic phone the study at any time without having to give a reason. The interviews with Voices of the 5% Study participants. These date of the withdrawal and any reasons given for withdrawal, were conducted by smartphones using a recorder applica- if provided, were recorded in the study notes. Recruitment tion. Due to Apple deleting the previously inbuilt phone ran from July 2020 to February 2021. call recording app, interviews conducted via iPhone used If participants returned a signed informed consent form the speaker function on the phone while holding an Evistr they were enrolled into the study and a baseline interview recorder close to the iPhone. was conducted. 2.6.3 Vowel Sustained Phonation 2.4 Vocal Biomarker Materials In a recording exercise performed after the initial recording, The audio recorder package provided to each participant con- speciﬁcally for this study, the recorders were returned to par- sisted of an Evistr Voice Recorder (model L157), earphones, ticipants in the same manner. Participants were instructed to USB cable, user manual, instruction sheet, and script to be emit the /a/ sound in a sustained manner after reading the read. Rainbow Passage once. Vowel phonation was recorded by 123 Acoustics Australia Fig. 1 Data collection ﬂowchart phone for six participants whose Evistr-recorded audio sam- 2.7 Vocal Parameters ples were not returned in time for analysis. Phone-recorded audio samples from interviews and sus- The set of vocal parameters measured were fundamental fre- tained vowel were all analysed as part of the phone-recorded quency, jitter, shimmer, noise, and formants. data set. 123 Acoustics Australia Fig. 2 Screenshot of PRAAT software: sound waveform (top), spectrogram (bottom). For the highlighted cycle, Period Length T 0.004478 s, F0 223.321 2.7.1 Fundamental Frequency (F0) 2.7.2 Jitter The fundamental frequency (F0) of a voiced sound is the fre- Jitter is the amount of variation in period length of the fun- quency, or number of oscillations per second, at which the damental frequency (illustrated in Fig. 3). Jitter parameters vocal cords vibrate when a human produces a sound (illus- assess perturbation in the frequency of the sound signal trated in Fig. 2). It is often referred to as pitch, which is our and indicate vocal deterioration caused by a lack of control perception of the fundamental frequency, and F0 is expressed of vocal fold vibration , an abnormality that has been in Hertz (Hz). Fundamental frequency ranges between 80 and observed during phonation in people who smoke . Jitter 450 Hz, is typically lower for males than females, and ﬂuc- values have been found to be higher in people who smoke tuates within speech for the same individual. [7, 8, 15]. Cigarette smoking has been observed to have an impact on We measured four different parameters of this variation: fundamental frequency. Speciﬁcally, the value of this acous- its absolute period-to-period difference (jitter absolute, jitta) tic parameter decreases [7, 8, 12] as a result of the effect of and three mean period-to-period difference scores: relative smoking on vocal cords. Gonzalez and Carpi  observed period-to-period variability of the pitch period (jitt), period- that this parameter was especially affected by smoking in to-period relative average perturbation (rap), and the pitch women and among people who had smoked for 10 years or perturbation quotient within 5 periods (ppq5). less. We anticipated observing low fundamental frequency in 2.7.3 Shimmer participants with a long history of smoking. Shimmer is the amount of variation in amplitude of the fun- damental frequency (illustrated in Fig. 3). Shimmer assesses 123 Acoustics Australia Fig. 3 Sound waveform: three cycles of fundamental frequency, with amplitude (shimmer) and period length (jitter) highlighted for one cycle perturbation in the amplitude of consecutive periods and indi- 3 Data Analysis cates vocal deterioration caused by lesions in the vocal cords . Amplitude abnormalities were observed during phona- 3.1 Sound Signal Investigated: /a/ tion in the voice of people who smoke , and shimmer values have been found to be higher in people who smoke Vowels are voiced sounds, and they are released with an [7, 8]. acoustic energy that is not restricted by articulation, unlike We measured four different parameters of this variation: what happens for many voiced consonants. Based on these its cycle-to-cycle difference in decibels (shdB) and three characteristics, vowels are the ideal segments for assessing mean cycle-to-cycle difference scores: relative evaluation of voice quality and measuring acoustic parameters relating to cycle-to-cycle variability of amplitude (shim), amplitude per- the vocal tract. turbation quotient over three periods (apq3), and amplitude We used Version 6.1.42 of the Praat software  to anal- perturbation quotient over ﬁve periods (apq5). yse the vocal parameters under investigation in this study: fundamental frequency (F0), jitter, shimmer, formants (F3 and F4), and noise ratio (NHR and HNR). 2.7.4 Noise ratio Studies by Zealouk et al.  and Pinto et al.  both selected the /a/ sound for their analysis of vocal parame- The noise-to-harmonic ratio (NHR) detects the presence of ters. Zealouk et al. extracted tokens of the vowel sound from noise in the sound signal, and it is measured in decibels (dB). recorded speech, while Pinto et al. analysed the sustained A higher NHR indicates worse voice quality. Current smok- emission of the /a/ sound by their participants. Both stud- ing is associated with an increase in the NHR . ies provide us with a reference point for this speciﬁc sound. No association has been documented between cigarette Additionally, similar to Zealouk et al., we extracted the /a/ smoking and a change in the value of the harmonic-to-noise sound from recorded naturalistic speech, and then, like Pinto ratio (HNR), which reﬂects the efﬁciency of speech. Accord- et al., we analysed the sustained phonation of /a/ recorded by ing to Teixeira et al. , “a value of less than 7 dB in HNR the participants. is considered pathological”. We were interested in HNR because it has been measured 3.2 Segments Extraction from the Rainbow Passage in a previous study assessing the effect of vaping on voice quality . Tuhanioglu ˘ et al.  found that mean HNR For each recording of participants reading the Rainbow Pas- values among people in the e-cigarette group and the con- trol group of people with no history of smoking were higher sage, we extracted tokens of the /a/ sound segments from the word ‘path’ contained in the following paragraph (emphasis than the mean HNR value for people who smoked. It was worth monitoring HNR values in our study because some of added): When the sunlight strikes raindrops in the air, they act as our participants reported switching to e-cigarettes during the study. a prism and form a rainbow. The rainbow is a division of white light into many beautiful colours. These take the shape of a long round arch, with its path high above, and its two 2.7.5 Formants ends apparently beyond the horizon. In NZ English, /a/ in the word ‘path’ tends to be released Formants indicate the resonant frequencies of the vocal tract. as a long vowel, which allowed for taking measures in For vowels, this information is relayed in four formants: the central part of the vowel, thus avoiding inﬂuence of F1, F2, F3, and F4 (illustrated in Fig. 4). Zealouk et al.  co-articulation from surrounding sound segments. Further- observed little correlation between current smoking and val- more, ‘path’ is a common word, on which participants ues of F1 and F2. However, F3 and F4 values for the vowels they investigated (/a/, /i/, and /u/) were lower in people who smoke. Therefore, we monitored the values of the two last However, it is possible to analyse connected speech; this was done formants, F3 and F4. with Cepstral peak prominence value in Murton et al. . 123 Acoustics Australia Fig. 4 Red lines on the spectrograph of sound [æ] indicate the four formants for this sound in this recording, from Formant 4 (F4), top line, to Formant 1 (F1), bottom line. The formant values at cursor point (34.18), about midway in the highlighted sound signal, are given in the window on the right. We are interested in the values of F3and F4, respectively, 2606.04 Hz and 4730.15 Hz, in this example were unlikely to stumble. Also, the word occurs some dis- 3.5 Parameters Analysis tance into the recorded speech when the speaker’s voice and speech delivery had stabilised. The vowel in the word ‘path’ 3.5.1 Fundamental Frequency F0 may be released as [a:], [a], or [æ], a variability that may present a limitation in any study that surveys sound segments Fundamental frequency analysis was controlled for extracted from natural speech. This variability, expected to sex/gender for three reasons. Firstly, because fundamental be participant-speciﬁc, and therefore carried over time in frequency was expected to be higher for women than men. subsequent recordings, would, however, be reﬂected in for- Secondly, because this frequency is especially affected in mants F1 and F2, which were not used in our measures, since women, even among people who had smoked for 10 years or they had not been revealed to be affected meaningfully by less . Thirdly, two of our participants were non-binary and cigarette smoking . at different stages in following a hormonal treatment that is likely to affect their vocal fundamental frequency. 3.3 Segments Extracted from Interviews 3.5.2 Jitter and Shimmer For each of the participants who provided a usable Evistr recording, a matching /a/ sound segment in the word ‘car’, Jitter parameters assess perturbation in the frequency of a or ‘start’, was isolated in their audio-recorded Voices of the sound signal, while shimmer values assess perturbation in its 5% baseline interview. The same vocal parameters were anal- amplitude. Smoking increases jitter and shimmer values. ysed, and then, the Evistr and Voices interview samples were compared. 3.5.3 Noise 3.4 Analysis of Sustained /a/ Sound Smoking is associated with an increase in the noise-to- Participants were asked to record themselves emitting a sus- harmonics ratio (NHR), which detects the presence of noise tained /a/ sound using the Evistr recording device; for six in the sound signal. The harmonic-to-noise ratio (HNR) participants, the sustained sound was phone-recorded. reﬂects the efﬁciency of speech and increases with vaping. 123 Acoustics Australia 3.5.4 Formants F3 and F4 or ‘start’. This happened when there was loss of sound signal during the phone interview, or because the partici- Similar to the fundamental frequency, the values of F3 and pant’s answer was very concise. Two participants provided F4 formant frequencies were expected to be lower for men no recordings of the vowel sustained phonation, and for six than women, and the presence of smoking has been observed participants, the recording was made via smartphone, rather to decrease their values. than with the Evistr audio recorder. Table 2 displays the mean and standard deviation (SD) values of the vocal parameters for the /a/ sound for the three types of audio samples. 3.6 Comparison of the Recording Type Table 3 displays the difference in means and statistical signiﬁcance (p value), by pair of audio samples (Table 3). The R function aov was used to compute the variance Table 4 displays the mean values of the vocal parame- between the means for the three types of audio samples, and ters for the audio signal recorded by smartphone or audio between the means for the different types of recording gear. recorder, the difference in means between the two types of Pairwise comparison (function TukeyHD) was used to deter- recording equipment, and the statistical signiﬁcance of this mine whether means between the three pairs of audio sample difference. and means between phone-recorded and Evistr-recorded sig- Furthermore, audio samples provided for a comparison of nals were statistically signiﬁcant. In this analysis, the p value the means of the vocal parameters measured for vowel phona- measures the probability that the difference in means could tion recorded via smartphone (six participants) and Evistr have occurred by chance, the lower the p value, the greater recorder (23 participants). The differences in means showed the statistical signiﬁcance of the mean difference. that phone-recorded audio samples displayed (a) highly sig- niﬁcant lower values for formants F3 (difference − 829; p value < 0.001) and F4 (difference − 1020; p value < 4 Results 0.001) and (b) signiﬁcant higher shim% value (difference + 2.73; p value < 0.05) and shim apq5 value (difference + 4.1 Response Rate 1.60; p value < 0.05). Means for the other vocal parameters showed no signiﬁcant difference. Of a total of 49 participants who were sent a recorder, 31 participants returned their initial recordings. Two participants withdrew from the study and 16 did not return their recorder 4.3.1 Fundamental and Formants Frequencies by courier or it arrived too late to be included in this analysis. The three tables below display mean and standard devia- tion values for the fundamental and formant frequencies by 4.2 Participant Characteristics sex/gender, for recorded readings (Table 5), phone interviews (Table 6), and sustained phonation (Table 7). Participant demographics are summarised in Table 1. 4.3 Recording Results 5 Discussion All but one participant recorded two readings of the Rain- bow Passage. The ﬁnal 61 readings, all captured on Evistr The aim of this study was to determine if (a) audio samples recorders, provided sufﬁcient material to calculate the mean of sufﬁcient quality could be collected at a distance for anal- and standard deviation values of the target vocal parameters ysis in Praat software and (b) these audio samples could be from the extracted sound signal. collected from naturalistic speech via smartphone or other The phone interviews of three participants provided no digital platforms. The study also aims to identify the opti- usable tokens of the /a/ sound segment in the words ‘car’ mum voice parameters to assess voice quality. Table 1 Participant Participants Number Age range Mean age Number, Number, Mean years smoking demographics at baseline age ≤ 40 age > 40 Women 16 24–63 36 12 4 19.5 (5–45) Men 13 19–81 48 6 7 31.5 (2–63.5) Non-binary 2 26–30 28 2 0 13 (10.5–15) All 31 19–81 40.5 20 11 24 (2–63.5) 123 Acoustics Australia 5.1 Comparing Audio Samples Reported values for the vocal parameters we surveyed show a gradation for the three audio samples. The sustained phona- tion audio samples were of the highest quality, with values consistently tending toward those of healthy voices . In contrast, phone interview samples return values that are the closest to those found in people who smoke [7–9]. In the middle, the measures obtained for recorded readings range between those of the two other audio samples. This gradation is remarkably consistent across parameters, and the difference in means is highly signiﬁcant between sustained phonation and interviews. Sustained phonation samples show (a) the highest frequency values (F0 and for- mants F3 and F4) and speech efﬁciency (harmonic-to-noise ratio), (b) the lowest frequency perturbation (jitter, shimmer), and (c) the presence of noise in the sound signal (NHR). This pattern is reversed for phone interview samples. As for recorded readings, values obtained for formants F3 and F4 show no signiﬁcant difference with those obtained for sustained phonation, while F0 and jitter values show no signiﬁcant difference with values observed for interviews. Shimmer and NHR values sit midway between values of the two other audio samples, and HNR is closer to the audio samples extracted from the interviews. The acoustic characteristics of the sustained phonation of the /a/ sound gave overall measures that correlate with voices healthier than in the two other audio samples. This is proba- bly due to the controlled emission of the participant’s voice recorded in this sample, which departs from the naturalis- tic speech segments extracted from the two other samples. These other samples are affected by the variability that typi- cally occurs in naturalistic speech. 5.2 Effect of Recording Equipment The measures of vocal parameters analysed by recording equipment suggest that sound signals recorded by smart- phones amplify the perturbations detected in acoustic char- acteristics of voice. In comparison with Evistr recordings, the overall measures for smartphone-recorded signals show (a) a pattern of signiﬁcantly lower F3, F4, and HNR val- ues and (b) signiﬁcantly higher shimmer and NHR values. Used as a one-off, these measures may erroneously point to an unhealthy voice, but they could be used in a longitudinal study to track variability in acoustic characteristics over time. 5.2.1 Type of Phone Recording Device and Application The sound quality of the participants’ voice was uneven across phone-recorded interviews. Common problems included crackling or ﬂattening, inaudible voice, background noise, and loss of sound signal. Twelve interviews displayed Table 2 Vocal parameters for sound signal /a/ analysed in baseline readings, interviews, and sustained phonation Vocal parameters F0 jitt jitta rap ppq5 shim shdB apq3 apq5 NHR HNR F3 F4 (%) (µs) (%) (%) (%) (dB) (%) (%) Recorded readings—participants n 31, analysed tokens n 61, recorded: EVISTR audio recorder, date range Nov 2020–March 2021 Mean 160 1.10 78.70 0.49 0.54 6.11 0.57 2.71 3.37 0.12 12.85 2653 3660 SD 42.5 0.84 68.00 0.45 0.41 3.12 0.28 1.56 1.85 0.09 3.88 248 244 Interviews—participants n 28, analysed tokens n 66, recorded: smartphone + recording application, date range July 2020–April 2021 Mean 151 1.49 110.85 0.58 0.64 8.58 0.76 4.36 * 0.23 10.00 1974 2826 SD 43.3 1.11 84.07 0.65 0.46 2.99 0.26 2.12 * 0.20 3.75 364 407 Sustained phonation—participants n 29, analysed tokens n 29, recorded: EVISTR audio recorder**, date range June 2021–August 2021 Mean 170 0.54 34.60 0.27 0.30 4.07 0.37 2.12 2.47 0.05 17.42 2599 3689 SD 43.7 0.44 28.10 0.23 0.23 2.99 0.27 1.55 1.81 0.05 4.58 494 539 *Shimmer apq5% values were indeterminate for the phone-recorded audio samples of several participants’ interview **For 6 participants, sustained /a/ was recorded with a smartphone using a recording application Acoustics Australia better sound quality than the other interviews, although our overall results include the measures of vocal parameters for all 28 interviews (as mentioned above, the phone interviews of three participants provided no usable tokens of the /a/ sound). It was not a goal of this study to compare sound signals recorded by different types of smartphones (refer to Jannetts et al.  for such comparison), but rather to compare sound signals (a) captured by audio recorder and smartphone and (b) extracted from different types of audio samples. We were able to select the segments that presented the best sound quality in all interviews, and the measures of the vocal parameters extracted from good sound-quality interviews did not depart signiﬁcantly from the other interviews. The interviews that produced good sound quality were made with (a) iPhone smartphones using the TapeACall© application or(b) Android-operated smartphones using the CubeACR© call recorder application . 5.2.2 Comparisons with Previous Studies How did our measures of vocal parameters compare with val- ues observed in previous studies? Vocal parameters extracted from the three audio samples had mean values consistent with previous surveys of groups of people who smoked[7–9]. For jitter and shimmer parameters, which measure perturbation in the period and amplitude of the fundamental frequency, these values were higher than those documented in previous studies (e.g. Pinto et al. ). With the exception of the fun- damental frequency measured in our male participants, and probably owing to the prevalence of participants under the age of 40, overall mean values all depart from values docu- mented for people who do not smoke and/or healthy voices in previous studies . 5.3 Fundamental Frequency Smoking decreases fundamental frequency values. There- fore, we expected to observe lower fundamental frequencies among our participants than in people who do not smoke. This is consistent with results reported in Pinto et al.  and Zealouk et al.  studies which compared mean fun- damental frequencies in groups of adults who smoked and who did not smoke. However, the overall fundamental fre- quency means that we reported, for all three audio samples, were consistently higher than for the studies [7, 8]. Further- more, the mean value for the fundamental frequency from our male participants was even higher than the mean value for the group of people who did not smoke in Pinto et al.  (114.49 Hz). The overall mean values for the fundamental Cepstral peak prominence  was not measured in these previous studies. This parameter is also useful at detecting aspects of poor voice quality and could be added in the future studies. Table 3 Vocal parameters: Comparison of means and statistical signiﬁcance, by pair of audio samples Vocal parameters F0 jitt jitta rap ppq5 shim shdB apq3 apq5 NHR HNR F3 F4 (%) (µs) (%) (%) (%) (dB) (%) (%) Recorded readings—interviews Means difference 8.80 − 0.38 − 32.10 − 0.08 − 0.10 − 2.46 − 0.19 − 1.64 * − 0.10 2.84 679 834 p Value 0.71 0.18 0.14 0.75 0.52 < 0.01 < 0.05 < 0.05 * < 0.01 < 0.05 < 0.001 < 0.001 Recorded readings—sustained phonation Means difference − 9.53 0.55 44.19 0.21 0.23 2.04 0.18 0.58 0.89 0.07 − 4.58 54 − 29 p Value 0.66 < 0.05 < 0.05 0.18 < 0.05 < 0.05 < 0.05 0.4 0.32 0.08 < 0.01 0.84 0.95 Interviews—sustained phonation Means difference − 18.3 0.94 76.29 0.30 0.34 4.50 0.38 2.23 * 0.18 − 7.42 − 624 − 863 p Value 0.24 < 0.001 < 0.001 < 0.05 < 0.01 < 0.001 < 0.001 < 0.001 * < 0.001 < 0.001 < 0.001 < 0.001 *Shimmer apq5% values were indeterminate for the audio samples of several participants’ interview Acoustics Australia frequency measured in our study (160.09 Hz for recorded readings and 169.63 Hz for sustained phonation) were also nearer to the overall value of the group of participants who did not smoke in Zealouk et al.  (168 Hz), than to that of their group of participants who smoked (143 Hz). The difference in the mean age of the participants may be the reason for this conﬂict, since age is also a factor in the value of this vocal parameter, which decreases with age. Two-thirds of our participants were 40 years old or under and had been smoking for less than 20 years. As a result, they had been exposed to fewer smoking years than subjects in previous studies. For example, adults who had been smoking for less than 20 years were excluded from Pinto et al.’s study, whose participants all ranged between 40 and 60 years of age. Young individuals are less likely to suffer from the effect of age on vocal cords and therefore present higher fundamental frequencies than people in old age. 5.4 Formants F3and F4 Zealouk et al.’s study  returned mean values circa 2500 for F3 and 3500 for F4, for a group of adults who were smoking at the time of the survey—consisting of 20 men, aged 28–50 with a mean age of 39, and most of them having smoked for at least 13 years. In our study, the overall mean values that we obtained across the three types of recordings were slightly higher for F3 and F4, but well under the mean values reported by Zealouk et al. for their control group of people who did not smoke: 3000 for F3 and 4000 for F4. Furthermore, as was mentioned earlier in the discussion, in our study formants F3 and F4 values were even lower when voice was recorded with a smartphone. 5.5 Jitter and Shimmer The mean jitter and shimmer values found in our study were high in comparison with the mean values obtained for healthy voices by Teixeira and Fernandes . All three sets of values in our study show some perturbation in the frequency and amplitude of the sound signal. The overall mean values for the /a/ sound that we logged for jitter and shimmer parameters in our study were higher than the values reported by Pinto et al.  for participants who were smoking, with the notable exception of the jitter values that we logged for the sustained phonation of the /a/ sound signal, which were lower than those reported in Pinto et al.’s study. Jitter and shimmer values were also higher than the values documented by Zealouk et al.  for jitter parameters, and on a par with the values they reported for shimmer parameters, although once again the sound phonation departed from this pattern and showed lower values than in Zealouk et al.’s study for jitta, ppq5, shdb, and apq5. Table 4 Measures of vocal parameters analysed by recording equipment Vocal parameters F0 jitt jitta rap ppq5 shim shdB apq3 apq5 NHR HNR F3 F4 (%) (µs) (%) (%) (%) (dB) (%) (%) Audio recorder: n 54 Mean 164 0.84 58.8 0.38 0.42 5.00 0.46 2.34 2.84 0.08 14.8 2703 3762 SD 43 0.72 58.4 0.37 0.34 3.21 0.29 1.56 1.86 0.08 4.72 303 321 Phone recorded: n 34 Mean 155 1.36 99.2 0.55 0.60 8.16 0.72 4.15 * 0.20 11.2 1968 2835 SD 44 1.09 81.8 0.61 0.46 3.11 0.27 2.07 * 0.19 4.83 361 380 Means difference (phone means—recorder means) − 9 0.52 40.4 0.17 0.18 3.16 0.26 1.81 * 0.12 − 3.6 − 735 − 927 p Value 0.374 < 0.01 < 0.01 0.124 < 0.05 < 0.001 < 0.001 < 0.001 * < 0.001 < 0.001 < 0.001 < 0.001 *Shimmer apq5% values were indeterminate for the audio samples of several participants’ phone recorded interview Acoustics Australia Table 5 Recorded readings: vocal frequency values by gender 5.6 Noise Vocal frequencies F0 F3 F4 NHR mean values (0.05–0.23) were equal or higher than the mean of 0.05 (SD: 0.15) documented by Pinto et al. for Women their group of people who smoked. Our participants’ voices Mean 189.18 2764.15 3661.47 presented noise in the sound signal and a worse voice quality SD 28.72 270.96 337.53 than the group surveyed in . Men Our participants’ mean HNR values, ranging from 10 to Mean 123.34 2512.44 3618.18 17.42 across the three types of recordings, were much lower SD 27.93 214.94 242.54 than the 23.9 HNR mean obtained for healthy voices by Non-binary Teixeira and Fernandes , indicating an overall lower efﬁ- Mean 157.08 2621.46 3921.79 ciency of speech in our participants, across all three types of SD 51.07 586.35 828.39 audio samples. 6 Conclusions Table 6 Phone interviews: vocal frequency values by gender Vocal frequencies F0 F3 F4 Cigarette smoking continues to be the leading preventable cause of death of about 7 million people globally per annum Women . Many of these deaths occur unnecessarily and pre- Mean 170.55 1854.75 2630.99 maturely. Earlier diagnosis of smoking-related harm, and SD 35.59 341.64 252.04 subsequent earlier intervention to support abstinence from Men smoking, could help reduce morbidity and years of life lost Mean 132.91 2009.95 2868.70 due to smoking. The vocal features surveyed in this study SD 41.96 337.37 321.54 have been associated with a clinical outcome, and these biomarkers  may be useful for early diagnosis or moni- Non-binary toring of progression of smoking-related diseases. Mean 145.56 2514.82 3809.05 This study was limited by the small sample size. A fur- SD 70.96 126.92 168.59 ther limitation is that a minor number of voice recordings on the smartphone were recorded with an external device rather than a smartphone application. The audio for these may have suffered additional noise and signal distortion. Nevertheless, Table 7 Sustained phonation: vocal frequency values by gender the results can be used to inform the value of pursuing more costly research and development of diagnostic voice tech- Vocal frequencies F0 F3 F4 nologies. It is useful to know that the task of self-completing Women an at-home voice recording was acceptable to a demograph- Mean 195.01 2681.50 3830.91 ically diverse group of adults who smoked. Secondly, the recordings were of sufﬁcient quality for measuring vocal SD 28.14 533.52 444.88 biomarkers using Praat voice analysis software. Men Finding that sound signals from spontaneous speech (read- Mean 134.28 2548.46 3540.83 ing and conversation) can be extracted and usefully analysed SD 41.24 367.70 651.58 is important for advancing the voice diagnosis ﬁeld. This Non-binary study contributes to that advancement by demonstrating that Mean 160.96 2213.92 3374.51 audio samples extracted from smartphone-recorded inter- SD 20.41 882.76 386.75 views in a naturalistic real-world setting were of sufﬁcient quality to be used for acoustic analysis in Praat. However, our results show that audio signals recorded with a smartphone display signiﬁcant perturbation, in comparison with signals Frequency of the sound signals also appeared to be recorded with an audio recorder. The devices can be cali- affected by the use of electronic communication devices and brated to account for systematic errors, the important thing shimmer parameters values were signiﬁcantly increased for being that the error remains the same and the random error sound signals recorded with smartphones. is not too high. 123 Acoustics Australia The results have implications for the advancement of source, provide a link to the Creative Commons licence, and indi- cate if changes were made. The images or other third party material remote delivery of health monitoring and for clinical tri- in this article are included in the article’s Creative Commons licence, als collecting data remotely. The relatively lower cost of unless indicated otherwise in a credit line to the material. If material collecting vocal biomarker data by smartphone or other dig- is not included in the article’s Creative Commons licence and your ital platforms could be particularly useful for addressing intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copy- inequities in health diagnosis and cessation support experi- right holder. To view a copy of this licence, visit http://creativecomm enced by people living in rural or underserviced areas, such ons.org/licenses/by/4.0/. as in low-income countries. Future studies could assess the potential of providing individuals who smoke with feedback on the health of their voice as a tool to enhance risk per- ceptions and motivate cessation . However, this study suggests that such interventions should focus on people who References have smoked heavily and persistently over decades. 1. Decoster, W., Debruyne, F.: Longitudinal voice changes: facts and This study also provides useful information for the design interpretation. J. Voice 14(2), 184–193 (2000) of future voice diagnostic research. For instance, sustained 2. Banjara, H., Mungutwar, V., Singh, D., Gupta, A.: Objective and vowel phonation would provide less intra-participant vari- subjective evaluation of larynx in smokers and nonsmokers: a ability and may be the optimum parameter for analysis of comparative study. Indian J. Otolaryngol. Head Neck Surg. 66(1), 99–109 (2014). https://doi.org/10.1007/s12070-011-0342-3 vocal changes among participants in longitudinal studies who 3. Guimarães, I., Abberton, E.: Health and voice quality in smokers: overtime may continue to smoke, may switch to vaping or an exploratory investigation. Logoped. Phoniatr. Vocol. 30(3–4), may stop smoking. 185–191 (2005). https://doi.org/10.1080/14015430500294114 4. Deng, Y., Wang, M., Zhou, L., Zheng, Y., Li, N., Tian, T., Zhai, Acknowledgements We are grateful to the Voices of the 5% Study Z., Yang, S., Hao, Q., Wu, Y., Song, D., Zhang, D., Lyu, J., Dai, participants for without them this research would not have been possible. Z.: Global burden of larynx cancer, 1990–2017: estimates from We are thankful also to Voices of the 5% Study staff and interviewers, the global burden of disease 2017 study. Aging 12(3), 2545–2583 in particular Emma Hurrell and Karen Potter. (2020). https://doi.org/10.18632/aging.102762 5. New Zealand Cancer Registry, Ministry of Health: Cancer: New Author contributions MG conceived of the study and obtained funding Registrations and deaths—series (2018). https://www.health.govt. for the study. MFD conducted the data extraction and PRAAT analysis. nz/nz-health-statistics/health-statistics-and-data-sets/cancer-new- All authors contributed to searching the literature, analysis, writing and registrations-and-deaths-series. Accessed 19 July 2021 editing of the paper. All authors approved of the ﬁnal version. 6. Hudson, M., Milne, M., Reynolds, P., Russell, K., & Smith, B.: Te ara tika guidelines for Maori ¯ research ethics: a framework for researchers and ethics committee members. Health Research Coun- Funding Funding was provided by Foundation for a Smoke-Free cil of New Zealand (2010) World (Grant No. FSFW COE1-009). 7. Pinto, A.G., Crespo, A.N., Mourão, L.F.: Inﬂuence of smoking isolated and associated to multifactorial aspects in vocal acoustic parameters. Braz. J. Otorhinolaryngol. 80(1), 60–67 (2014). https:// Declarations doi.org/10.5935/1808-8694.20140013 8. Zealouk, O., Satori, H., Hamidi, M., Laaidi, N., Satori, K.: Vocal parameters analysis of smoker using Amazigh language. Int. J. Competing interests This study was funded with a grant from the Speech Technol. 21(1), 85–91 (2018). https://doi.org/10.1007/ Foundation for a Smoke-Free World, Inc. (“FSFW”), a US nonproﬁt s10772-017-9487-0 501(c)(3) private foundation with a mission to end smoking in this 9. Gonzalez, J., Carpi, A.: Early effects of smoking on the voice: a generation. This study is, under the terms of the grant agreement (COE1- multidimensional study. Med. Sci. Monit. 10(12), 649–656 (2004) 009) with FSFW, editorially independent of FSFW. The contents, 10. Boersma, P., Weenink, D.: Praat: doing phonetics by computer. selection and presentation of facts, as well as any opinions expressed Version 6.1.42 (2021). http://www.praat.org/ herein, are the sole responsibility of the authors and under no circum- 11. Fujiki, R.B., Chapleau, A., Sundarrajan, A., McKenna, V., stances should they be regarded as reﬂecting the positions of FSFW. Sivasankar, M.P.: The interaction of surface hydration and vocal None of the authors and nor does the Centre have any commercial loading on voice measures. J. Voice 31(2), 211–217 (2017). https:// interests in any nicotine or tobacco products. doi.org/10.1016/j.jvoice.2016.07.005 12. Byeon, H., Cha, S.: Evaluating the effects of smoking on the voice Ethics approval Ethics approval was sought from the Health and Dis- and subjective voice problems using a meta-analysis approach. ability Ethics Committee (HDEC), but the study was deemed out of Sci. Rep. 10(1), 4720 (2020). https://doi.org/10.1038/s41598-020- scope as it was an observational study that would not involve more than 61565-3 minimal risk, that is, no more risk than participants might encounter 13. Teixeira, J.P., Fernandes, P.O.: Jitter, Shimmer and HNR classiﬁ- during everyday life. cation within gender, tones and vowels in healthy voices. Procedia Technol. 16, 1228–1237 (2014). https://doi.org/10.1016/j.protcy. Open Access This article is licensed under a Creative Commons 2014.10.138 Attribution 4.0 International License, which permits use, sharing, adap- 14. Pinar, D., Cincik, H., Erkul, E., Gungor, A.: Investigating the effects tation, distribution and reproduction in any medium or format, as of smoking on young adult male voice by using multidimensional long as you give appropriate credit to the original author(s) and the methods. J. Voice 30(6), 721–725 (2016). https://doi.org/10.1016/ j.jvoice.2015.07.007 123 Acoustics Australia 15. Teixeira, J.P., Oliveira, C., Lopes, C.: Vocal acoustic analy- 19. Epic Enterprises LLC. TapeACall: Call Recorder (2019) sis—Jitter, Shimmer and HNR parameters. Procedia Technol. 9, 20. Cube Apps Limited. Cube ACR (2019) 1112–1122 (2013). https://doi.org/10.1016/j.protcy.2013.12.124 21. World Health Organisation, Tobacco (2020). https://www.who.int/ 16. Tuhanioglu, ˘ B., Erkan, S.O., Özdas, ¸ T., Derici, Ç., Tüzün, K., news-room/fact-sheets/detail/tobacco/. Accessed 14 July 2021 Senkal, ¸ Ö.A.: The effect of electronic cigarettes on voice quality. 22. Fagherazzi, G., Fischer, A., Ismael, M., Despotovic, V.: Voice for J. Voice 33(5), 811.e13-811.e17 (2019). https://doi.org/10.1016/j. health: the use of vocal biomarkers from research to clinical prac- jvoice.2018.03.015 tice. Digit. Biomark. 5(1), 78–88 (2021). https://doi.org/10.1159/ 17. Murton, O., Hillman, R., Mehta, D.: Cepstral peak prominence 000515346 values for clinical voice evaluation. Am. J. Speech Lang. Pathol. 23. Young, R.P., Hopkins, R.J.: Increasing smokers’ risk percep- 29(3), 1596–1607 (2020). https://doi.org/10.1044/2020_AJSLP- tion improves CT screening participation. Thorax 67(9), 834–835 20-00001 (2011). https://doi.org/10.1136/thoraxjnl-2011-201453 18. Jannetts, S., Schaefﬂer, F., Beck, J., Cowen, S.: Assessing voice health using smartphones: bias and random error of acoustic voice parameters captured by different smartphone types. Int. J. Lang. Publisher’s Note Springer Nature remains neutral with regard to juris- Commun. Disord. 54(2), 292–305 (2019). https://doi.org/10.1111/ dictional claims in published maps and institutional afﬁliations. 1460-6984.12457
Acoustics Australia – Springer Journals
Published: Mar 1, 2023
Keywords: Voice recording; Acoustic analysis; Vocal biomarkers; Larynx; Smoking
Access the full text.
Sign up today, get DeepDyve free for 14 days.