Chronic obstructive pulmonary disease (COPD) is one of the leading causes of death worldwide. To manage the increasing number of COPD patients and reduce the social and economic burden of treatment, healthcare providers have sought to implement remote patient monitoring (RPM). Screen-based RPM applications, such as filling self-reports on the smartphone or computer, have been shown to increase the quality of life, reduce the frequency and severity of exacerbations, and increase physical activity in patients with COPD. These applications, however, are not without challenges for the elderly target population. They are often used on devices designed by and for a diferent age group, which makes filling out self-reports prone to error and induces fears of technology malfunctions. Voice-based conversational agents (VCAs) are available on more than 2.5 billion devices and are increasingly present in homes worldwide. Aside from their commercial success, VCAs are also credited with several functionalities, such as hands-free use, that make their adoption in healthcare attractive, especially for the elderly. In this work, we investigate the potential of VCAs for RPM of COPD. Specifically, we designed and evaluated Lena, a single-board computer-based VCA framed as a digital member of the medical team. Lena acts as RPM for the early prediction of COPD exacerbations by asking ten symptom-related questions to determine the patient's daily health status. This paper presents the patients' feedback after their interaction with Lena. Patients evaluated the acceptability of the system. Notably, all patients could imagine using the system once a day in the context of a larger study and wished to integrate Lena into their daily routine.
M
inflammation due to smoking or long–term exposure to pollutants (e.g., dust, fumes, poor air quality) [2].
and to reduce the pressure on health services, providers have sought to introduce RPM for COPD patients [3].
COPD, physiological parameters of COPD patients are not continuously monitored outside of hospitals, except for research purposes [10]. Moreover, symptoms are usually self–reported by the patients using pen and paper diaries [10]. Considering that COPD patients belong to the older part of the population, current screen–based applications (e.g., filling self–reports on smartphones or computers) do not come without challenges. Older adults often have low IT–literacy, fear of malfunction [ 11], and a lack of confidence in their ICT abilities [ 12]. In addition, the fact that most commercially available software is developed by and for a diferent age group limits the inclusion of more digital applications in the lives of older people, even though they recognize the benefits that percent of all global deaths [2]. The disease’s prevalence come with increased ICT capabilities [13]. Further, there is on the rise due to higher smoking prevalence and the is evidence that older generations prefer voice–based to aging population [2]. In consequence, the burden on screen–based communication [14]. health care providers is expected to increase in the comFrom The Voder, the first attempt to electronically syn- ing years. Evidence suggests that RPM for COPD patients thesize human speech by using a console with fiteen can reduce hospitalizations [6] and costs associated with touch–sensitive keys and a pedal to select the appro- this disabling disease by at least 14 percent [7]. Moreover, priate bandpass filters to convert the hisses and tones McLean et al. showed in a Cochrane review that RPM can into vowels, unveiled at the 1939 New York World’s Fair, increase the quality of life of COPD patients and reduce to today’s Alexa, Siri, or Google Assistant, the way and the number of exacerbations [8]. There is also evidence in frequency humans interact with VCAs have evolved dra- a review by Lundell et al. that RPM can improve physical matically. VCAs are now available on more than 2.5 activity levels in COPD patients [9]. Pedone and Lelli, in billion devices such as smartphones and tablets, smart their review [24], reported a positive but nonsignificant speakers, computers and have even been embedded in efect of remote care on hospital admissions and emerwearables or cars, thus nearly becoming ubiquitous in gency department visits. Similarly, McDowell et al. [25] our daily lives [15]. One in four Americans owns a smart showed that RPM with self–management improves the speaker, and in 2018 alone, their ownership increased quality of life but does not significantly improve emerby 40 percent [16]. Beyond their commercial success, gency care.
VCAs are also credited with several functionalities that More recently, Rassouli et al. investigated the association make their use in healthcare attractive. VCAs enable between the COPD Assessment Test (CAT) and the risk hands–free interaction, allow input from individuals with of exacerbation of COPD [26]. In this study, patients comlow literacy or with intellectual [17], motor and cogni- pleted an online questionnaire focused on detecting an tive disabilities [18], or provide more natural support acute exacerbation of COPD (daily) and the CAT (weekly). for routine health tasks when in–person healthcare is The authors found that by completing the questionnaires not possible [19]. Voice interaction also enables passive regularly, patients could assess their health status more monitoring and analysis of audio samples for healthcare accurately. Also, the evolution of the CAT could help applications, such as Alzheimer’s [20], depression [21], assess the risk of future exacerbations. and schizophrenia [22]. In addition, recent work suggested the potential of speech (e.g., pause time, pause 2.2. VCAs for healthcare frequency, prosodic features, among others) as a marker of exacerbations in patients with COPD [23]. Humans apply social rules in interactions with computWith this in mind, we argue that VCAs have the potential ers [27, 28]. VCAs enable this behavior by mimicking to enable RPM for COPD patients by facilitating the col- interpersonal conversation. Therefore, compared to text– lection and sharing of health–related information with based interaction, they are perceived as more socially healthcare professionals, thereby improving quality of present (i.e., the perception of interacting with an inlife, reducing exacerbations, and thus, reducing the costs telligent being) [29, 30] and even more believable than in COPD care. humans when it comes to information retrieval [31]. SimFor these reasons, the authors developed and evaluated a ilar to existing screen–based conversational agents [32, single–board computer (SBC)–based VCA framed as Lena, 33, 34], VCAs can form a working alliance [35] with paa digital member of the medical team. This pilot study tients, which has a positive impact on treatment outinvestigates the acceptability of a voice–based approach comes [36, 37]. Recent works by Balasuriya et al. [38], to alleviating patients’ communication burden while fill- Ireland et al. [39], and Kadariya et al. [40] also suggest ing questionnaires. More specifically, the VCA’s ability that non–commercial dedicated VCAs can meet user exto formulate its questions simply and understandably, pectations when supporting the prevention and manas well as its ability to understand the patient’s answer agement of chronic or mental illnesses. Furthermore, and respond accordingly, were assessed with four COPD speech interaction also enables passive monitoring and patients. analysis of voice samples for health applications, such as Alzheimer’s [20], depression [21], schizophrenia [22], and COPD [23].
3.2.2. How Lena communicates tinuously improved and eventually integrated into one To engage with a patient, Lena must be able to (i) recogsystem. nize when the patient talks with it, (ii) understand the patient’s speech, and (iii) to respond in an intelligible way. 3.1. Hardware The next three paragraphs describe the implementation of these points.
Board Computer (SBC), a USB microphone, and a pair of Key phrase recognition To activate Lena, a patient USB speakers. For aesthetics and protection, a wooded must utter the key phrase “Hello, Lena”. The developed loudspeaker housing contains Lena’s core components, app recognizes this key phrase in real–time. After recas shown in Figure 1. More precisely, Lena’s core consists ognizing the key phrase, Lena starts speaking, explains of a Raspberry Pi 4 Model B (cf. Figure 1. D.; Specifica- the conversation’s goal to the patient, and asks the first tions: 8 GB RAM and 1.5 GHz processor) and a SanDisk question.
Ultra microSDHC of 16 GB. The USB microphone has The key phrase recognition uses Vosk, an ofline open– a sensitivity of −67 dBV/pBar, −47 dBV/Pascal ±4 dB, source speech recognition toolkit that understands 17 the frequency ranges from 100 − 16kHz, and on–device languages and dialects. noise–canceling filters out the background noise. Finally, the USB–powered speakers are connected to the 3.5 mm jack port of the SBC. Their frequency ranges from 50 to 20′000 Hz.
3.2.1. Operating system
and open–source operating system based on the Android mobile platform. This has the advantage that all apps already available in the Google Play Store can also be used on the Raspberry Pi after installing Open GApps. The Open GApps Project is an open–source efort that provides pre–built Google Apps packages. By installing Open GApps, the Google Play Store became available, as well as Google’s speech functionalities and APIs, which will be discussed in more detail below.
tient’s responses, Lena uses Android’s API. More specifically, the SpeechRecognizer, RecognitionService, and RecognizerIntent Android speech classes handle the speech– to–text transformation of the patient’s speech. We implemented speech recognition to operate purely ofline.
In this way, the patient’s personal and sensitive data remains on the device and does not need to be anonymized by another algorithm, which could afect the quality of the recording. To enable this functionality, the authors had to specify the flag EXTRA_PREFER_OFFLINE in the Android RecognizerIntent class. Furthermore, the ofline speech recognition package must be downloaded on the device from the Language and Input section of the LineageOS system settings. Finally, the Android speech recognition framework must be set to the one provided by Google.
Speech synthesis Lena’s ofline text–to–speech (TTS) Lena would react diferently, e.g., by telling a joke if the capabilities are also based on the Android API. The An- patient was not feeling well. At the beginning of the droid class TextToSpeech converts Lena’s predefined con- interview, Lena would ask the patient whether to use versational turns from text to speech. Other TTS APIs formal or casual language throughout the interview. Fi(e.g., Amazon Polly, Google Cloud TTS, IBM Watson nally, the eight other binary questions were aimed at TTS) could have been used to make Lena’s speech more evaluating the patient’s health status. melodic and less monotone. However, the authors de- At the end of the dialog, Lena would thank the patient cided to use the Android API, which allows speech to be for sharing today’s symptoms with the study team and synthesized ofline without installing additional software switch back into an idle mode, waiting for the patient to packages while delivering an intelligible speech. say the key phrase.
3.3.2. Evaluation and interview details 3.3.1. Patient recruitment & interaction with the After the interaction with Lena, the patients were interVCA viewed by the study team for 20 minutes.
The first part consisted of filling out a pen and paper Four COPD patients (three male, one female) of age 69 ± 5 survey with six questions evaluated on a 7–point Likert interacted with Lena. In practice, Lena was framed as a scale (see Table 1). The questions aimed to evaluate Lena digital member of the medical team. and were based on the technology acceptance model [42]. Three diferent hospital rooms served to conduct the in- Perceived enjoyment was defined as the degree to which terviews (a furnished hospital bedroom, a doctor’s ofice, the activity of using technology is perceived to be enand an examination room) to evaluate Lena under var- joyable [43]. Finally, relative advantage represented the ious circumstances. Also, one patient received oxygen degree to which a novel application is perceived as being therapy while interacting with Lena. Before the interac- better than the state of the art [44]. The second part tion with Lena began, the authors instructed the patients of the interview consisted of a face–to–face interview on the nature of the questions that would be asked. The between the patient and a member of the study team (see patients were randomly recruited based on their previous Table 2). or current participation in a COPD–RPM study [41, 26].
In this study, patients completed a daily questionnaire via their personal computer or smartphone. Rassouli et al. 4. Results were able to detect 60 out of 63 acute exacerbations of COPD with a sensitivity, specificity, positive and negative 4.1. Acceptability predictive value of 95, 98, 26, and 99.9 %, respectively [41].
During the interaction, Lena asked and the patients answered the following questions in German: (i) Do you have more dyspnoea today, exceeding your usual variation?, (ii) Do you have more sputum today, exceeding your usual variation?, (iii) Is your sputum today more yellow or green, exceeding your usual variation?, (iv) Do you have more cough today, exceeding your usual variation?, (v) Do you feel febrile today?, (vi) Do you feel like having a common cold today?, (vii) Do you feel unwell today, exceeding your usual variation? and (viii) Did you start to take your emergency medication within the last 24 h?. This time, however, the questions were asked and answered verbally. By selecting these patients, it was possible to objectively compare the speech–based solution with the previously used screen–based solution.
Concretely, the patients sat in front of Lena, as shown in Figure 1. After being triggered with the key phrase “Hello Lena”, Lena began asking questions to capture a questionnaire of the patient’s perceptions of current COPD severity. The questionnaire consisted of nine binary questions and one open–ended question. The latter referred to the patient’s mood. Depending on the patient’s mood,
and felt that Lena understood their responses (EOU2).
They enjoyed conversing with Lena (ENJ1) and could imagine using Lena at least once a day to complete the questionnaire (ITU1), and found Lena useful (USE1). Last but not least, patients would prefer to use Lena than the existing screen–based solution (RA1). The entire conversation lasted an average of two minutes.
After the conversation with Lena, the authors conducted face–to–face interviews with patients to understand the current system’s strengths and weaknesses. Table 2 shows a condensed summary of these interviews. All patients owned a smartphone (TECH1). Half of the patients were already intentionally using a VCA in the context of driving or reading a text aloud (TECH2). All patients could imagine using Lena instead of the screen–based application for a one–year study (ITU2). When asked about the advantages and disadvantages of completing the questionnaire orally, patients emphasized the ease of use of Lena and that no login is required, which is the case with the screen–based study (ENJ3). Another patient responded that interacting verbally with Lena would increase compliance as it felt more engaging (ENJ3). Two shortcomings were that Lena sometimes took too long each interview in JSON format. The patients could freely to interpret her utterances, and one of the two testers answer Lena’s questions. However, to understand the highlighted Lena’s intonation as a disadvantage (IMP1). positive or negative connotation of a patient’s answer, In their opinion, these two factors made the dialogue the algorithm searched for a keyword such as “yes” or sluggish at some moments. One patient was concerned “no”. that Lena would not fit in his luggage when traveling All the interviews could be initiated with the key phrase on vacation, and that in this case, completing the ques- and completed by the patients. It took, on average, 2±1.41 tionnaire would require a diferent solution (IMP2). P4 attempts to trigger the interaction. Finally, the patient’s also pointed out possible risks caused by power outages speech transcribed by Lena perfectly matched the ausince the system needs to be plugged in (IMP2). Finally, thors’ transcription; all responses were correctly identiwhether Lena should rather address the patients in Swiss ifed and transcribed.
German dialect instead of German was answered positively by two patients (ENJ4).
ing, and interacting with Lena to complete the questionnaire. The speech recognition results indicate that the prototype can understand the patients’ responses with perfect accuracy and lead the discussion accordingly.
Also, all patients declared their willingness to use Lena symptoms and make information exchange seamless with daily over a longer period (e.g., 12 months) (ITU1, ITU2) the medical team. The combination of the voice–based and favored this voice–based solution over the original questionnaires and the passively recorded data should screen–based application [41] (RA1). This suggests that not only open a wide range of new research directions Lena qualifies for validation in a longitudinal study with but, more importantly, provide better support for COPD COPD patients. patients.
Lena’s voice assistant capabilities rely on Android’s speech recognition and speech synthesis and Vosk’s speech recognition APIs. All three can be used ofline without trans- 6. Conclusion ferring patients’ recorded speech samples to an external server. With this approach, we recognize the sensitive This pilot study proposes Lena, a state of the art VCA nature of medical and speech data and ensure privacy for COPD RPM. Lena interacts with the patient in a spoand security. ken natural language to collect daily self–reports. This Although the healthcare sector already uses VCAs [45, ifrst evaluation yielded promising acceptance results of 46], this study provides the first insights regarding the a VCA–based RPM application for COPD. All patients feasibility, relevance, and acceptability for such an appli- also showed a willingness to integrate Lena into their cation for COPD patients. In contrast to the proposed daily routine and saw its potential to improve future RPM pen and paper or screen–based applications to collect solutions. We plan to integrate and evaluate Lena in a self–reports’ information, Lena provides a tailored ap- longitudinal observational study with COPD patients. plication for the elderly target population by verbally This research is a first step towards enabling scalable and capturing patient information without requiring interac- natural–language–delivered RPM to facilitate access to tion with another person via a phone call or in–person health–related self–management services. It may also visit [41, 24, 9]. Moreover, recent work has suggested help overcome limitations of text–based systems, such the potential of speech (e.g., pause time, pause frequency, as the lack of literacy of users in countries with low eduprosodic features, a.s.o.) and cough [47] as a marker cation index, accessibility for the elderly population, or of COPD exacerbations [23]. Considering that recent even empowerment of patients with mental, motor, or research has also shown the feasibility to detect cough cognitive disabilities. events with high accuracy on devices with limited computing power (e.g., smartphones [48]) and the ability to References continuously listen in the background, detect and count coughs comparable to human annotators [49], we argue [1] T. G. I. for Chronic Obstructive Lung Disease, 2020 that such a smart speaker system could not only collect GOLD REPORTS, 2020. URL: https://goldcopd.org/ subjective information but also provide passively and gold-reports/. objectively measured parameters related to the health [2] WHO, Chronic respiratory diseases, 2017. URL: status of patients with COPD. https://bit.ly/39y6qQN.
The study population received Lena well in terms of use- [3] C. E. Bolton, C. S. Waters, S. Peirce, G. Elwyn, fulness, ease and intention of use, and enjoyment. Never- Insuficient evidence of benefit: a systematic retheless, there is a need for a systematic and comparative view of home telemonitoring for copd, J Eval evaluation of the system (e.g., test the distance and direc- Clin Pract 17 (2011) 1216–22. doi:1 0 . 1 1 1 1 / j . 1 3 6 5 tion in which the microphone was efective, test on older 2 7 5 3 . 2 0 1 0 . 0 1 5 3 6 . x . adults with an absence of basic digital literacy skills) to [4] I. Tomasic, N. Tomasic, R. Trobec, M. Krpan, understand which aspects of Lena are best for patient T. Kelava, Continuous remote monitoring of acceptance and better outcomes than standard in–person copd patients—justification and explanation of care. the requirements and a survey of the available Patients welcomed completing the questionnaire without technologies, Medical & Biological Engineering having to sign in. However, for Lena to meet the same & Computing 56 (2018) 547–569. URL: https:// data security standards as the state of the art screen– dx.doi.org/10.1007/s11517-018-1798-z. doi:1 0 . 1 0 0 7 / based system, voice–based authentication could be used, s 1 1 5 1 7 - 0 1 8 - 1 7 9 8 - z . as it prevents unauthorized use and protects patient pri- [5] S. D. Sullivan, S. D. Ramsey, T. A. Lee, The vacy [38]. economic burden of copd, Chest 117 (2000) Finally, Lena’s current version did not allow patients to 5S – 9S. URL: http://www.sciencedirect.com/ freely express their symptoms since the questionnaire science/article/pii/S0012369215527487. doi:h t t p s : consisted only of binary questions. Nonetheless, using / / d o i . o r g / 1 0 . 1 3 7 8 / c h e s t . 1 1 7 . 2 _ s u p p l . 5 S . a voice–based system allows patients to communicate [6] F. C. Cordova, D. Ciccolella, C. Grabianowski, more naturally to provide more information about their J. Gaughan, K. Brennan, F. Goldstein, M. R. Jacobs, G. J. Criner, A telemedicine-based intervention among low income housing residents in new york reduces the frequency and severity of copd exacer- city, BMC Research Notes 10 (2017) 517. URL: https: bation symptoms: A randomized, controlled trial, //doi.org/10.1186/s13104-017-2850-z. doi:1 0 . 1 1 8 6 / Telemed J E Health 22 (2016) 114–122. doi:1 0 . 1 0 8 9 / s 1 3 1 0 4 - 0 1 7 - 2 8 5 0 - z .
t m j . 2 0 1 5 . 0 0 3 5 . [15] B. Kinsella, A. Mutchler, Voice assistant consumer [7] G. Paré, P. Poba-Nzaou, C. Sicotte, A. Beaupré, adoption in healthcare, Technical Report, 2019.
E. Lefrançois, D. Nault, D. Saint-Jules, Com- URL: https://voicebot.ai/voice-assistant-consumerparing the costs of home telemonitoring and adoption-report-for-healthcare-2019. usual care of chronic obstructive pulmonary [16] B. Kinsella, A. Mutchler, Smart Speaker disease patients: A randomized controlled Consumer Adoption Report, Technitrial, European Research in Telemedicine / cal Report March, 2019. URL: https: La Recherche Européenne en Télémédecine 2 //voicebot.ai/wp-content/uploads/2019/03/ (2013) 35–47. URL: http://www.sciencedirect.com/ smart_speaker_consumer_adoption_report_2019.pdf. science/article/pii/S2212764X13000241. doi:h t t p s : [17] S. S. Balasuriya, L. Sitbon, A. A. Bayor, / / d o i . o r g / 1 0 . 1 0 1 6 / j . e u r t e l . 2 0 1 3 . 0 5 . 0 0 1 . M. Hoogstrate, M. Brereton, Use of voice [8] S. McLean, U. Nurmatov, J. L. Liu, C. Pagliari, activated interfaces by people with intellectual J. Car, A. Sheikh, Telehealthcare for chronic disability, in: Proceedings of the 30th Australian obstructive pulmonary disease, Cochrane Conference on Computer-Human Interaction, Database Syst Rev (2011) Cd007718. doi:1 0 . 1 0 0 2 / OzCHI ’18, Association for Computing Ma1 4 6 5 1 8 5 8 . C D 0 0 7 7 1 8 . p u b 2 . chinery, New York, NY, USA, 2018, p. 102–112. [9] S. Lundell, A. Holmner, B. Rehn, A. Nyberg, URL: https://doi.org/10.1145/3292147.3292161.
K. Wadell, Telehealthcare in copd: a system- doi:1 0 . 1 1 4 5 / 3 2 9 2 1 4 7 . 3 2 9 2 1 6 1 . atic review and meta-analysis on physical out- [18] F. Masina, V. Orso, P. Pluchino, G. Dainese, S. Volcomes and dyspnea, Respir Med 109 (2015) 11–26. pato, C. Nelini, D. Mapelli, A. Spagnolli, L. Gamdoi:1 0 . 1 0 1 6 / j . r m e d . 2 0 1 4 . 1 0 . 0 0 8 . berini, Investigating the accessibility of voice [10] F. Pitta, T. Troosters, V. S. Probst, M. A. Spruit, assistants with impaired users: Mixed methods M. Decramer, R. Gosselink, Quantifying physical study, Journal of Medical Internet Research activity in daily life with questionnaires and mo- 22 (2020) e18431. URL: https://dx.doi.org/10.2196/ tion sensors in copd, Eur Respir J 27 (2006) 1040–55. 18431. doi:1 0 . 2 1 9 6 / 1 8 4 3 1 .
doi:1 0 . 1 1 8 3 / 0 9 0 3 1 9 3 6 . 0 6 . 0 0 0 6 4 1 0 5 . [19] E. Sezgin, L. K. Militello, Y. Huang, S. Lin, A scop[11] J. Kowalski, A. Jaskulska, K. Skorupska, K. Abram- ing review of patient-facing, behavioral health inczuk, C. Biele, W. Kopeć, K. Marasek, Older terventions with voice assistant technology taradults and voice interaction: A pilot study with geting self-management and healthy lifestyle begoogle home, in: Extended Abstracts of the 2019 haviors, Translational Behavioral Medicine 10 CHI Conference on Human Factors in Comput- (2020) 606–628. URL: https://dx.doi.org/10.1093/ ing Systems, CHI EA ’19, Association for Com- tbm/ibz141. doi:1 0 . 1 0 9 3 / t b m / i b z 1 4 1 . puting Machinery, New York, NY, USA, 2019, p. [20] M. L. B. Pulido, J. B. A. Hernández, M. A. F. 1–6. URL: https://doi.org/10.1145/3290607.3312973. Ballester, C. M. T. González, J. Mekyska, doi:1 0 . 1 1 4 5 / 3 2 9 0 6 0 7 . 3 3 1 2 9 7 3 . Z. Smékal, Alzheimer’s disease and auto[12] W. Kopeć, R. Nielek, A. Wierzbicki, Guide- matic speech analysis: A review, Expert Systems lines towards better participation of older adults with Applications 150 (2020) 113213. URL: in software development processes using a new https://dx.doi.org/10.1016/j.eswa.2020.113213. spiral method and participatory approach, in: doi:1 0 . 1 0 1 6 / j . e s w a . 2 0 2 0 . 1 1 3 2 1 3 .
Proceedings of the 11th International Workshop [21] J. Wang, L. Zhang, T. Liu, W. Pan, B. Hu, T. Zhu, on Cooperative and Human Aspects of Soft- Acoustic diferences between healthy and depressed ware Engineering, CHASE ’18, Association for people: a cross-situation study, BMC Psychiatry 19 Computing Machinery, New York, NY, USA, (2019). URL: https://dx.doi.org/10.1186/s12888-0192018, p. 49–56. URL: https://doi.org/10.1145/ 2300-7. doi:1 0 . 1 1 8 6 / s 1 2 8 8 8 - 0 1 9 - 2 3 0 0 - 7 . 3195836.3195840. doi:1 0 . 1 1 4 5 / 3 1 9 5 8 3 6 . 3 1 9 5 8 4 0 . [22] N. B. Mota, M. Copelli, S. Ribeiro, Thought disor[13] A. Aula, Learning to use Computers at a Later Age, der measured as random speech structure classifies in: HCI and the Older Population, Univ. of Glasgow, negative symptoms and schizophrenia diagnosis Leeds, UK, 2004. 6 months in advance, npj Schizophrenia 3 (2017). [14] D. T. Duncan, W. C. Goedel, J. H. Williams, B. Elbel, URL: https://dx.doi.org/10.1038/s41537-017-0019-3.
Acceptability of smartphone text- and voice-based doi:1 0 . 1 0 3 8 / s 4 1 5 3 7 - 0 1 7 - 0 0 1 9 - 3 . ecological momentary assessment (ema) methods [23] V. Nathan, K. Vatanparvar, M. M. Rahman, E. Neand intrinsic motivation to use computers in the workplace 1, Journal of applied social psychology 22 (1992) 1111–1132. [44] E. M. Rogers, Difusion of Innovations, 5th ed., Free
Press, 2003. [45] Aiva Health, Virtual health assistant | aiva health,
2020. URL: https://www.aivahealth.com/. [46] Mayo Clinic, Skills from mayo clinic, 2020. URL:
https://www.mayoclinic.org/voice/apps. [47] M. G. Crooks, A. den Brinker, Y. Hayman, J. D.
Williamson, A. Innes, C. E. Wright, P. Hill, A. H.
Morice, Continuous cough monitoring using ambient sound recording during convalescence from a copd exacerbation, Lung 195 (2017) 289–294. URL: https://doi.org/10.1007/s00408-0179996-2. doi:1 0 . 1 0 0 7 / s 0 0 4 0 8 - 0 1 7 - 9 9 9 6 - 2 . [48] F. Barata, K. Kipfer, M. Weber, P. Tinschert,
E. Fleisch, T. Kowatsch, Towards device-agnostic mobile cough detection with convolutional neural networks, in: 2019 IEEE International Conference on Healthcare Informatics (ICHI), 2019, pp. 1–11.
doi:1 0 . 1 1 0 9 / I C H I . 2 0 1 9 . 8 9 0 4 5 5 4 . [49] F. Barata, P. Tinschert, F. Rassouli, C. Steurer-Stey,
E. Fleisch, M. A. Puhan, M. Brutsche, D. Kotz, T. Kowatsch, Automatic recognition, segmentation, and sex assignment of nocturnal asthmatic coughs and cough epochs in smartphone audio recordings: Observational field study, Journal of Medical Internet Research 22 (2020) e18082. URL: https: //dx.doi.org/10.2196/18082. doi:1 0 . 2 1 9 6 / 1 8 0 8 2 .