Effect of an iOS App on Voice Therapy Adherence and Motivation

Abstract

Purpose

Patients commonly report difficulties adhering to voice therapy. An iOS app was developed in our lab that assists practice via reminder notifications, instructional recordings, and cepstral peak prominence analysis results. The purpose of this study was to assess the effect of such homework support modality on adherence behavior and associated motivation in a comparison of app support and written homework instructions and to assess the usability and utility of the app.

Method

Thirty-four individuals exhibiting adducted hyperfunction were randomized to receive either written homework instructions or the app when practicing resonant voice exercises for 3 weeks. All patients digitally audio-recorded all home practice, provided self-reported estimates of generalization, and completed weekly motivation scales.

Results

App support significantly increased practice frequency but did not affect self-reported generalization or motivation. Practice was significantly predicted by System Usability Scale scores. Utility of reminders and instructions were good, but cepstral peak prominence feedback was considered useful to only a subset of participants.

Conclusion

Interactive mobile therapy support can significantly increase practice of resonant voice homework without influencing motivation.

Adherence is defined by the World Health Organization as “the extent to which a behavior—taking medication, following a diet, or executing a lifestyle change—corresponds with agreed-upon clinician recommendations” (Sabaté, 2003). Between weekly therapy sessions, direct voice therapy asks patients to replace problematic voice production behaviors with more optimal target techniques such as resonant, confidential, or loud voice (Abbott, 2008; Ramig et al., 2004; Ramig & Verdolini, 1998; Roy et al., 2003; Verdolini-Marston et al., 1995; Ziegler et al., 2010). To achieve this goal, homework may include (a) daily practice of the target voice production technique at a variety of levels (e.g., sound, syllable, word, sentence, and paragraph) and (b) the intentional implementation (i.e., conscious generalization) of this target technique in daily communication such as conversation and teaching (Abbott, 2008; Gartner-Schmidt et al., 2016; Ramig et al., 2004; Roy et al., 2003). Recommended practice frequencies typically range from at least two practice sessions per day for resonant voice (Abbott, 2008) to seven sessions of conversational practice in conversational training therapy (Gartner-Schmidt et al., 2016; Gillespie et al., 2019). Thus, across approaches, treatment adherence involves substantial extra clinical patient engagement.

As in all behavioral interventions, adherence is a problem in voice therapy (Behrman et al., 2008; Hapner et al., 2009). Patients practice less frequently than recommended (van Leer & Connor, 2010, 2012, 2015), struggle to apply the target technique in daily communication (van Leer & Connor, 2010; Van Stan et al., 2017), drop out of therapy altogether (Hapner et al., 2009), or relapse to a pretherapy vocal status after successful discharge (Van Lierde et al., 2007), perhaps due to inadequate long-term adherence to target voice use. Thus, improving adherence is important for optimizing voice therapy outcomes and maintenance.

When provided with written homework instructions in resonant voice therapy, patients commonly report a variety of self-regulatory and motor learning barriers to completing this homework. These include difficulty with (a) remembering to practice, (b) remembering to use resonant voice in daily communication, (c) recalling and reproducing resonant voice technique independently without the clinician's model, and (d) self-evaluating the accuracy of their resonant voice production without the clinician's corrective feedback (van Leer & Connor, 2010). Patients state that failure to overcome these adherence barriers can halt practice until the next therapy session and cause treatment drop out if failure persists. Therefore, adherence support tools for resonant voice therapy should aim to reduce or resolve the common patient-perceived barriers to adherence.

Social Cognitive Factors in Adherence

Adherence behavior is fundamentally governed by the same principles of social cognitive theory (Bandura, 1989) that underlie all volitional, goal-directed human behaviors, such as the pursuit of health and medical goals (Rimer & Glanz, 2005), academic learning (Pajares, 1996), and professional skills in the workplace (Locke & Latham, 2002). Social cognitive theory views behavior as determined by three factors: (a) past behavior, including successes and failures in goal attainment; (b) external factors, such as environmental and social support; and (c) internal factors, such as patient beliefs and emotions (Bandura, 2001). These behavioral, external, and internal factors interact in a triadic fashion as illustrated in Figure 1.

Figure 1.

Social cognitive model of triadic asymmetrical reciprocal causation.

Within the category of internal factors, two mutable beliefs constitute the primary motivational determinants of health, academic, and occupational goal attainment: (a) self-efficacy, or people's confidence in their ability to attain a goal (Bandura, 2001), and (b) goal commitment: their determination to do so (Locke & Latham, 1990). These beliefs are considered part of a person's motivation to pursue a goal (Amrhein et al., 2003; Locke & Latham, 2002; W. R. Miller & Rollnick, 2012). Because these beliefs are mutable rather than trait-based, they can be enhanced to drive greater adherence. For example, self-efficacy for a task can be strengthened by observing one's own successful performance on video: the “self as model” (Bellini & Akullian, 2007; Prater et al., 2012). Self-efficacy can be garnered vicariously by observing the successful practice attempts of peers with whom one identifies: the “peer model” (Decker & Buggey, 2014; Wang et al., 2011). Likewise, goal commitment can be increased by (a) simplifying the targeted goal, (b) increasing goal concreteness so that its attainment is less abstract and ambiguous, or (c) providing feedback to illustrate progress toward the goal (Easthall et al., 2013; Locke & Latham, 1994). Thus, adherence behavior can be improved by enhancing internal motivational factors or increasing external support for the behavior. Since self-efficacy and goal commitment were found to predict a significant amount of practice variance in voice therapy (van Leer & Connor, 2015), they are worth targeting in an adherence intervention.

Mobile Interventions to Improve Adherence

Mobile devices have the potential to improve voice therapy adherence by providing various forms of adherence support anytime, anywhere. Within social cognitive theory's triadic model of reciprocal causation, the mobile intervention can be conceptualized as an external factor that either affects behavior directly as illustrated by the black arrow in Figure 2, or does so indirectly by enhancing motivation, as illustrated by the gray arrows. To improve both motivation for adherence and adherence behavior itself, a series of mobile adherence support approaches were developed and tested in our laboratory. These included the provision video and audio models of the target voice quality (van Leer & Connor, 2012, 2015; van Leer & Porcaro, 2019) to address patients' difficulty recalling and reproducing the target voice, as well as an interactive feedback app to help patients self-evaluate the accuracy of their target voice productions (van Leer et al., 2017). This study builds directly and iteratively on these previous studies, which will be discussed next.

Figure 2.

The role of mobile support in the triadic model.

MP4 Video Models Improve Adherence

Loaded onto MP4 players, instructional videos have been shown to significantly increase both resonant voice therapy adherence behavior as well as motivation for adherence. In a randomized crossover study, patients practiced significantly more frequently and had greater motivation for practice when they were provided with clinician-model and patient self-as-model MP4 videos than when these same patients were given written homework instructions (van Leer & Connor, 2012). Clinician videos were perceived as useful in clarifying homework content and also as emotionally supportive (“I felt like the therapist was with me all the time”). Self-as-model videos had a positive effect on self-efficacy by affirming “that I could do it, and it wasn't weird.” In a subsequent randomized trial comparing MP4 support (i.e., clinician, self, and peer videos) with written homework instructions, participants in the MP4 group reported significantly greater use of the target voice in daily communication than controls and significantly greater self-efficacy for doing so, without practicing more frequently in isolation (van Leer & Connor, 2015). MP4 participants also achieved greater Voice Handicap Index (VHI; Jacobson et al., 1997) and Consensus Auditory–Perceptual Evaluation of Voice (CAPE-V; Kempster et al., 2009) score reductions than controls, but these differences were not significant. Thus, the adherence intervention should be strengthened for greater effects.

App-Based Support

While MP4 devices assisted practice through provision of video models, they lacked the capacity to provide reminder notifications and to supply interactive feedback, defined as corrective information regarding accuracy of performance (Schmidt et al., 2018). Thus, the barriers surrounding remembering and self-evaluating resonant voice production were not addressed with MP4 technology. The development of mobile applications provided a technical platform for addressing these barriers.

Acoustic Analysis as Interactive Feedback

Since lack of feedback is reportedly a barrier to adherence, development of a patient-centered mobile feedback system has the potential to improve adherence. Mobile feedback regarding loudness levels has demonstrated potential for helping patients reach loudness goals (Cole et al., 2007; Halpern et al., 2012; Nolan, 2013; Van Stan et al., 2017), but mobile tools to support voice quality goals are in the early stages of development and feasibility testing.

With acoustic analysis results serving as a form of augmented feedback, display of the Multidimensional Voice Profile (Pentax Medical) has previously demonstrated utility in helping dysphonic patients discern the difference between their habitual and resonant voice at initial evaluation (Bonilha & Dawson, 2012) in a nonmobile context. The Multidimensional Voice Profile program calculates 33 acoustic parameters and displays 12 of these (e.g., jitter, shimmer, harmonics-to-noise ratio) within a circle representing normal limits for each value (Amir et al., 2007). Since patients' resonant voice was more normal along these parameters than their habitual voice, the results display was useful to patients (Bonilha & Dawson, 2012). Likewise, provision of fundamental frequency information, sound pressure level, and 3–5 kHz formant amplitude was found somewhat useful in training vocally healthy actors to produce a resonant voice (Laukkanen et al., 2004). Thus, a variety of acoustic measures have demonstrated utility as an augmented feedback approach for differentiating resonant from normal or dysphonic voice. However, these forms of feedback were studied only in the context of therapy or training sessions (i.e., with clinician guidance), as the acoustic analysis programs were neither mobile nor client centered. Thus, the effect on adherence and motivation for adherence were not examined.

Cepstral Peak Prominence as Mobile Feedback

To provide patients with a form of mobile feedback for differentiating habitual and resonant voice independently, a patient-centered iOS app was developed that conducts cepstral peak prominence (CPP) analysis at the touch of a button (van Leer et al., 2017). CPP was chosen for its strong association with the perceptual judgment of overall voice quality of dysphonic individuals (Stepp et al., 2011). Since resonant voice strategies yield an overall improved voice quality as early as the first session (Bonilha & Dawson, 2012; Dejonckere & Lebacq, 2001), CPP can be used to capture improved voice quality resulting from resonant voice probes throughout therapy. Likewise, in nondysphonic speakers, CPP has been found to differentiate habitual from trained resonant voice productions (Aydınlı et al., 2019). CPP has also demonstrated feasibility as ambulatory feedback tool in voice therapy for hyperfunctional voice patients (Huo, 2020). Furthermore, unlike perturbation measures, CPP is suitable for analysis of connected speech (Heman-Ackah et al., 2003; Watts & Awan, 2011) and thus holds greater flexibility and ecological validity as potential feedback tool. Thus, there are multiple reasons to consider CPP as feedback tool in resonant voice therapy.

When tested empirically in the clinic, CPP values in both sustained phonation and connected speech were significantly higher for resonant voice production than for habitual voice quality in a heterogeneous group of patients, demonstrating its utility for differentiating dysphonic from resonant voice (van Leer et al., 2017). When these patients were provided with the CPP app to aid their practice, they put forth greater effort to achieve resonant voice (e.g., standing up to take a deeper breath) than when practicing without it: a gamification effect. On average, they also reported significantly higher self-efficacy for practice in the CPP condition than in the unassisted practice condition (van Leer et al., 2017). Consistent with the goal setting theory's premise that concrete, measurable, objective feedback is inherently motivating (Locke & Latham, 1994), patients stated that “it was good to put a number” to their target voice and note the difference in CPP values between habitual and target voice production. As such, mobile CPP demonstrated potential to motivate resonant voice practice.

Reminder Notifications for Improved Adherence

Mobile reminder notifications have demonstrated efficacy for increasing adherence to a variety of health behaviors (Badawy & Kuhns, 2017; Kannisto et al., 2014; Vervloet, van Dijk, et al., 2012) but have not been tested for their utility or efficacy in increasing voice therapy practice. In the triadic model, mobile reminders can be conceptualized as external factors that trigger behaviors directly, rather than doing so via motivation. Given their efficacy across health behaviors, they are likely to assist voice therapy patients as well.

A Comprehensive Voice Therapy Adherence iOS App

In order to create a comprehensive mobile support system for voice therapy adherence, an iOS app was developed in our lab that combines the three tools described previously: (a) reminder notifications, (b) instructional recorded examples (clinician and self-as-model), and (c) CPP feedback. This app was hypothesized to exceed the efficacy of MP4 support in increasing adherence, because it contains reminders and CPP feedback in addition to models. The app also eliminates the use of self-report as a measure of practice frequency and duration because it digitally records all dedicated patient practice.

Study Purpose

Since the app was hypothesized to influence practice and associated patient motivation, the primary purpose of this study was threefold: (1) to examine the effect of homework modality (i.e., app support vs. written instructions) on resonant voice practice frequency and duration, as quantified via digital recording; (2) to examine the effect of the app on motivation for practice (i.e., self-efficacy and goal commitment); and (3) to evaluate the usability (i.e., ease of use) and utility (i.e., usefulness) of the app.

Participants all underwent four sessions of voice therapy but were randomized to receive either app-based or written homework support. The app group was hypothesized to outperform the control group on all adherence and motivation measures. Purpose 3 was exploratory rather than hypothesis driven, with the intent to assess participant perspectives on app features and uncover problems and goals for future development. Although the app did not include specific features to support or measure implementation of resonant voice in daily communication (i.e., intentional generalization), patient-perceived generalization and associated motivation were also examined for direct comparison with MP4 study results.

Given the initial and exploratory rather than confirmatory nature of this first investigation, experiment-wise error was not controlled, in order to reduce the probability of a Type 2 error masking clinically meaningful differences that may guide future work in adherence and app development.

Method

Research Design and Schedule

An experimental investigation was conducted into the effect of homework support modality on voice therapy adherence behavior and associated motivation for adherence to accomplish Purposes 1 and 2 of this study. The design was a prospective randomized trial of homework condition—not of treatment approach—comparing app-based homework support to written homework instructions over a four-session course of resonant voice therapy. Homework condition thus served as independent variable, while weekly adherence and motivation measures served as dependent variables. To accomplish Purpose 3 (assessment of app usability and utility), app group participants evaluated the usability and utility of the app formally at the end of the study period. Approval was granted by the Georgia State University Institutional Review Board.

Initial evaluation occurred within a week prior to Session 1 or immediately preceding it. Posttherapy evaluation was conducted at the conclusion of Session 4. Each of the four voice therapy sessions was 1 hr in duration and spaced 1 week apart, resulting in a total of 3 weeks of intersession home practice, as illustrated in Figure 3. Thus, the first week of practice took place between Therapy Sessions 1 and 2, the second week between Sessions 2 and 3, and the third and final week of practice between Sessions 3 and 4.

Figure 3.

Adherence schedule. Four sessions of therapy were separated by a total of 3 weeks of practice.

Participants

Participants were referred to the study directly by multiple otolaryngologists in the greater Atlanta metro area or were self-referred via flyers on the Georgia State University campus and subsequently examined by an otolaryngologist if stroboscopic screening revealed a concern. Treating physicians made the determination of laryngeal status, whereas the primary investigator determined inclusion based on functional status. Functional inclusion criteria were (a) laryngeal hyperfunction (Hillman et al., 1989; Scherer et al., 1987) as characterized by the perception of strained voice quality, (b) participant self-report of laryngeal effort or vocal fatigue, and (c) positive response (i.e., improved voice quality and reduced effort) to resonant voice therapy probes such as humming and other forward focus facilitators.

The study required individuals whose vocal production could be improved through resonant voice technique, so that resonant voice practice was a meaningful goal. Hence, primary inclusion criteria were stimulability for vocal improvement with resonant voice probes, presence of adducted hyperfunction (i.e., phonotraumatic voice production behavior), and report of increased vocal effort or vocal fatigue. Since voice production behavior was of primary importance, vocal fold status was allowed to vary, including both individuals with and without benign tissue changes. Three participants with diagnosis of presbyphonia were also included because they fit the inclusion criteria of adducted hyperfunction, self-reported vocal effort and fatigue, and stimulability to resonant voice probes. Further inclusion and exclusion criteria in Table 1 were chosen to limit the likelihood that progress or lack thereof would be caused by factors unrelated to treatment adherence.

Table 1.

Participant inclusion and exclusion criteria.

Inclusion criteria

Exclusion criteria

Adults ages 18–75 years Chronic (> 2 months) vocal complaints Stimulability for vocal improvement in response to resonant voice probes (e.g., humming) at initial evaluation Complaint of laryngeal effort Complaint of vocal fatigue Perception of strained voice production behavior consistent with adducted hyperfunction (Hillman et al., 1989) Vocal fold appearance: normal, age-related changes, changes associated with hyperfunction including midmembranous edema or benign lesions, hypervascularity, varices, vocal process granuloma, vocal fold scarring

Individuals ages < 18 or > 75 years Recent (< 2 months) history of vocal complaints Stimulability for normal voice in response to vegetative tasks but not in response to resonant voice probes Neurogenic voice or speech disorder Active tobacco use Undergoing surgery requiring intubation during the course of study enrollment Inability to manipulate mobile device

Of the 38 patients recruited, one was removed from the data set when she was not consistently responsive to resonant voice probes, sent for re-evaluation with her laryngologist, and found to have spasmodic dysphonia (an exclusion criterion for the study). Three participants dropped out (two controls and one app participant): The two controls were unable to meet study demands (i.e., attend sessions, record practice) with drop out occurring during Weeks 1 and 2, respectively; the app participant was lost to follow-up for no known reason during Week 3. Among the 34 remaining participants, 18 had been randomized to the control group (M _age = 36 years, SD = 17.8), including 14 women and four men, and 16 to the app group (M _age = 31.44 years, SD = 12.39), including 11 women and five men. Laryngeal status of participants is presented in Table 2. Treatment was provided by the first author and a graduate student in speech-language pathology.

Table 2.

Description of vocal fold characteristics for control and app groups.

Control group			App group
Age (years)	Sex	Vocal fold abnormality	Age (years)	Sex	Vocal fold abnormality
29	F	Nodules	37	F	Nodules
49	F	Varix/ectasia	29	F	Mild midmembranous edema
22	F	WNL	59	F	Nodules
29	M	WNL, mild R TVF edema	20	F	WNL
23	F	WNL	23	F	WNL, mild erythema
21	F	WNL	22	F	WNL, mild edema and erythema
23	F	Bilateral pseudocysts	24	M	Hypervascularity
64	F	Bilateral pseudosulcus	27	F	WNL, s/p thyroidectomy
32	F	Varix/ectasia	42	F	WNL, mild edema
25	M	Mild erythema	18	F	WNL
26	F	WNL, h/o LPR	37	M	Unilateral polyp
27	F	Nodules	30	M	Hypervascularity
58	F	WNL	31	M	Nodules
30	F	Mild midmembranous edema	21	F	Unilateral polyp
70	M	Vocal fold bowing (atrophy, age related)	57	F	Vocal fold bowing (atrophy, age related)
72	M	Vocal fold atrophy (atrophy, age related)	26	M	WNL, mild vocal fold edema, hypervascularity
26	F	Milld midmembranous edema
22	F	Mild midmembranous edema

Note. F = female; M = male; WNL = Within Normal Limits; R TVF= Right True vocal fold; s/p = status post; h/o = history; LPR = laryngo-pharyngeal reflux.

To control for effect of therapy approach, all participants attended 4-hr-long sessions of a resonant voice therapy approach (Abbott, 2008) as in our study evaluating the effect of MP4 support on adherence (van Leer & Connor, 2015). In the treatment hierarchy, resonant voice probes such as humming and voiced fricatives were used to elicit a salient resonant voice target followed by resonant voice production on a hierarchy of syllables, words, sentences, paragraphs, and conversation.

Independent Variable: Homework Condition

Homework goals included daily practice and implementation of resonant voice technique in daily communication. Specifically, both groups of participants were asked to (a) practice at least 3 times per day and for at least 5 min at a time or until benefit was perceived, consistent with the preceding randomized trial (van Leer & Connor, 2015) and resonant voice protocols (Abbott, 2008; Roy et al., 2001; K. Verdolini-Abbott, personal e-mail communication, August 20, 2018) and (b) implement resonant voice use in daily communication. The control group received individualized written homework instructions. The app group received homework instructions via the app, including (a) instructional audio recordings (i.e., clinician and self-as-model), (b) three daily reminders, and (c) CPP feedback.

Both groups were asked to digitally audio-record all homework practice for subsequent analysis in the lab. Control participants recorded their homework practice with the standard digital voice recorder app located on their own smartphone or on an iPod they borrowed from the lab. App participants recorded their practice while using the voice app designed for this study, which was likewise either downloaded onto their own iPhone or on an iPod borrowed from the lab.

App Procedures

At the start of the study, app participants received approximately 10 min of instruction in operating the app and additional follow-up instruction as needed on subsequent sessions. CPP instruction involved explanation of the measure itself, instruction on the use of the CPP functionality, and analysis of both habitual and target resonant voice production in sustained phonation and in a voiced resonant phrase of the participant's choosing (e.g., “Woo, woe, war” or “My oh my oh my”). Utility of CPP feedback was in differentiating participants habitual and target voice. In negative practice tasks, participants were trained to compare CPP values associated with their habitual voice (e.g., values of 21–22 dB) with those associated with their resonant voice production (e.g., 24–26 dB) to clarify the concrete difference between these modes of voice production. Thus, for each participant, CPP values for resonant voice were judged in relation to their habitual voice, not to an external standard. Participants were instructed to hold CPP tasks (i.e., vowels and phrases) constant for comparison between habitual and target voice and to hold mouth-to-mic distance constant at 20 cm. The app was programmed to lead the participant through a CPP test automatically at the start and end of each practice session so that participants could discern a potential change in value from the start of the session (i.e., habitual voice) to the end of the session (i.e., resonant voice). Participants could also access the CPP feature outside a programmed practice session, for their own reference, for example, in negative practice tasks.

Instructional audio recordings to support home practice were made during or at the end of each therapy session. These were composed of multiple brief (i.e., about 15–30 s) clinician instructions and demonstrations of individualized voice exercises and helpful cues, as well as participant self-as-model demonstrations of correct and incorrect (i.e., negative) practice. The app was programmed to prompt and record practice for a set duration (e.g., 30–120 s) following each instructional recording, in order to encourage practice and provide data for analysis in the lab. Complete exercise sets typically consisted of hierarchical progression of instructional recordings (e.g., target voice production in isolated sounds, syllables, phrases) each followed by about a minute of practice, for a grand total of 5 min of scheduled practice time.

After instructional recordings were saved to the app, three reminder notifications per day were scheduled via the app's reminder functionality. Reminder times were set in accordance with the participant's schedule; however, these occurred at the same time each day of the week due to calendar functionality limitations and did not vary by daily schedule differences.

Practice with the app commenced as follows. At the time of the scheduled reminder, an alert would sound, and a notification banner would appear on the screen. Swiping the notification banner would take the user to a start menu. Upon clicking on the start button, the user would be cued to take a CPP test on sustained phonation or on a connected speech task. Results would be displayed immediately on the screen after recording ended. An option was provided to repeat the CPP test as many times as desired if the participant was unhappy with their result. Next, the first exercise instruction played. A “skip” button was provided if instructions were deemed unnecessary for a review. Each instructional playback was followed by a preset period of recorded practice indicated by a circular countdown timer that would display on the screen. A skip button was provided in case the user wished to skip practice in part or altogether; partial completion of practice was saved for future analysis. At completion of timed practice, the user would be cued to continue to the next exercise or practice for an additional 30 s in order to achieve their target. When all exercises in a set were finished, a final CPP test was prompted for user comparison to pre-exercise values. Screen shots of the start menu, CPP functionality, exercise instructions and recorded practice, and homework completion screen are shown in Figure 4. Practice yielded digital time- and date-stamped .wav recordings for export and analysis in the laboratory.

Figure 4.

Voice therapy app screenshots. CPP = cepstral peak prominence.

Measures

Adherence to Practice

At the start of Therapy Sessions 2, 3, and 4, the past week's digital practice recordings were transferred from the participant's mobile device onto the laboratory computer research drive for analysis. For analysis, a “practice session” was defined as a single recording or series of consecutive recordings containing voice therapy homework practice. If a subsequent recording was initiated more than 30 min after the last, it was considered a new practice session rather than a continuation of the previous session. Per week, the number of practice sessions were documented in an Excel spreadsheet to yield weekly practice frequencies, and the duration of each session was noted in seconds.

Motivation for Practice

Two aspects of motivation—self-efficacy and goal commitment—were assessed in relation to practice. At the end of each treatment session, patients completed the Self-Efficacy Scale for Voice Therapy (van Leer & Connor, 2015) to capture their self-efficacy for practice in the upcoming week (see Table 3). Goal commitment for practice in the upcoming week was measured via Items 2 and 3 of the Readiness Ruler adapted for voice therapy (van Leer & Connor, 2015) from the original measure (W. R. Miller & Rollnick, 2012). These scales were chosen for their previous demonstration of predictive and convergent validity (van Leer & Connor, 2015) and to allow for comparison with previous work. Both scales use the same response format: an equal-appearing interval scale as response format, ranging from 0 (low) to 10 (high) and allowing mid values (e.g., 8.5), and guided by a 100-mm visual analog line

Table 3.

Voice therapy self-efficacy scale items.

Section 1: Practice How confident are you that you can practice: 0–10, N/A

1. When you have time to yourself (in the car) 2. When you are busy 3. When you are tired 4. When you are traveling (vacation, business) 5. When you don't have time alone 6. When other people can hear you practice 7. When people around you are unsupportive 8. When the exercises are silly 9. When you're not sure if you're practicing correctly 10. When you just don't feel like it

Section 2: Generalization How confident are you that you can use your target voice: 0–10, N/A

1. During voice therapy (in the voice clinic) 2. At work 3. In a professionally demanding situation 4. On the phone 5. In a loud environment 6. With people who are unsupportive of your voice problem 7. With your significant other 8. With your family 9. When you raise your voice or shout 10. When you are socializing 11. When you are under stress 12. When you are tired 13. When you are relaxed 14. When you are excited 15. When you can't concentrate on your voice 16. When people push your buttons 17. When you're talking to strangers (people who don't know you)

The Self-Efficacy Scale for Voice Therapy is composed of two sections. Section 1 asks participants to estimate their confidence in overcoming 10 barriers to practice. Section 2 asks participants to do so for 17 barriers to implementing the target voice in daily communication (i.e., intentional implementation). Section averages are calculated separately, yielding a self-efficacy score for practice and a separate score for intentional implementation in daily communication (van Leer & Connor, 2015).

The Readiness Ruler adapted to voice therapy is a three-item adaptation of the established Readiness Ruler Scale (W. R. Miller & Rollnick, 2012) and asks patients to assess their overall confidence in achieving a task in the upcoming week, their commitment to doing so, and its perceived importance in the context of other demands of the upcoming week (van Leer & Connor, 2015), as shown in Table 4. To measure goal commitment, Items 2 and 3 (i.e., commitment and importance) were averaged and analyzed.

Table 4.

Readiness ruler applied to voice therapy.

Voice Therapy Goal 1: Practice 3×/day: 0–10, N/A

How confident are you that you can achieve this goal? How committed are you to this goal? How important is this goal compared to other things you have to do this week?

Voice Therapy Goal 2: Stay in your best voice throughout the day: 0–10, N/A

How confident are you that you can achieve this goal? How committed are you to this goal? How important is this goal compared to other things you have to do this week?

Usability and Utility

At study conclusion, app participants completed the System Usability Scale (SUS; Brooke, 1996), a 10-item Likert-format scale that assesses the ease of use of a system. In addition, the app's utility (i.e., usefulness) in assisting practice was assessed qualitatively through a brief semistructured interview in which patients were asked what it was like to use the app and its individual features, what kinds of difficulties they encountered in using the app, and any suggestions they had for future development. Comments were transcribed and analyzed for main themes by the author and a research assistant. SUS scores were correlated with practice variables to determine the influence of perceived usability on aspects of practice.

Self-Perceived Generalization

Self-perceived generalization was assessed every session via self-report in the manner consistent with our previous research (van Leer & Connor, 2015; van Leer & Porcaro, 2019). Specific to each participant's target term (e.g., resonant, forward, buzzy), participants were asked, “What percent of time did you use this voice in the past week?” This question was also asked in the first therapy session, because some participants recognize that they used the target resonant technique some of the time at baseline. The question did not differentiate nonconscious generalization from intentional implementation of resonant voice.

Given that this measure has not been validated with objective measures of generalization (e.g., analysis of intermittent covert recordings) and that self-report is likely exaggerated (Kazantzis et al., 2004; van Leer & Porcaro, 2019), it is limited to a subjective appraisal of resonant voice use and can only be interpreted as such. However, since patients' subjective appraisal of progress plays a critical role in their attendance and dropout (Hapner et al., 2009; Portone-Maira et al., 2011; van Leer & Connor, 2010), it is worthy of documentation.

Motivation for Implementation of Resonant Voice

Motivation for implementing resonant voice technique was assessed in the same manner as motivation for regular practice: via the Readiness Ruler, Goal Commitment, and Self-Efficacy Scale for Voice Therapy. Thus, participants completed the same set of measures twice: one in reference to practice and once in reference to implementation. Self-efficacy for implementation was assessed via Section 2 of the Self-Efficacy Scale, which asks participants to rate their self-efficacy for applying the target voice in daily communication in the context of 17 barriers to doing so, each represented by a scale item.

Voice Outcome Measures

Although not the focus of this study, voice outcome measures were obtained at study onset and at the conclusion of Session 4 to assess group equivalence at baseline and to assure that both groups benefited from therapy. The study was not powered to detect group differences in treatment outcomes, and this was not its aim. However, measurement of outcome variable allowed for assessment of power for future studies.

In order to capture effects of vocal improvement on voice-related quality of life, participants completed the VHI (Jacobson et al., 1997). To capture voice quality improvement associated with resonant voice use, the CAPE-V (Kempster et al., 2009) was completed. CAPE-V tasks were recorded in a quiet room with a Roland RO5 digital .wav (Roland Corporation) and Glottal M80 headset microphone (Glottal Enterprises) placed at an 8-cm mouth-to-mic distance at a 45° angle.

At study completion, two speech-language pathologists with over 15 years of voice expertise served as perceptual raters. They were unfamiliar with participants and blinded to condition and pre- or posttherapy status. Participant recordings were played in random order at comfortable loudness over Sennheiser HD205 headphones (Sennheiser Corporation). Raters judged overall degree of dysphonia in accordance with the CAPE-V procedure of placing a hash mark on the 100-mm visual analog scale to indicate severity. To measure intrarater reliability, 10 samples were included twice, and to prevent perceptual drift, anchor recordings (Massachusetts Eye and Ear Database, Kay Pentax) were played every 20 recordings or as requested by the rater. Intrarater reliability was r = .86 for the first rater and r =.76 for the second rater, indicating acceptable reliability; interrater reliability was adequate (r =. 75), with one rater consistently rating samples more severe than the other resulting in reduced agreement.

Statistical Analysis

One-way analyses of variance (ANOVAs) were conducted to examine the effect of homework condition on practice frequency and duration. A sequential comparison of weekly practice frequency differences was conducted with t tests following the directional hypothesis that app support would exceed written homework support. Repeated-measures ANOVAs were conducted to examine within-group and between-groups effects for all remaining dependent variables, as these variables were measured repeatedly. All statistical analyses were performed with IBM SPSS statistics software Version 23. Experiment-wise error was not controlled in this study because a conservative approach could increase the probability of Type 2 errors and thus mask clinically meaningful differences that may guide future work.

Group Equivalence at Study Onset

For all measures taken at study onset, t tests were conducted to determine that there were no significant group differences, as shown in Table 5. On average, degree of dysphonia was mild. In addition, a chi-square test was conducted to examine group equivalence in the proportion of vocal fold abnormalities. The proportion of individuals with vocal fold abnormalities did not differ significantly by group, χ²(1, N = 34) = 1172, p = .732. Control participants did report substantially more use of the target voice at study onset than app participants, but this difference was not significant.

Table 5.

Group equivalence at treatment onset.

Measure	Control group M (SD)	App group M (SD)	t (df)	p
Age	36 (17.8)	32 (12.3)	0.857 (32)	.39
Voice Handicap Index (pretherapy)	36.11 (18.30)	33.19 (10.50)	0.492 (32)	.626
CAPE-V (pretherapy)	11.07 (12.6)	8.67 (7.2)	0.645 (31)	.524
Readiness Ruler Goal Commitment: Practice	8.44 (1.4)	8.48 (1.3)	0.093 (32)	.92
Readiness Ruler Goal Commitment: Generalization	6.61 (2.5)	6.96 (3.1)	0.337 (28)	.74
Self-Efficacy Scale: Practice	6.43 (2.1)	6.08 (1.7)	0.527 (32)	.602
Self-Efficacy Scale: Generalization	4.57 (3.1)	3.89 (3)	0.659 (32)	.515
Self-perceived generalization %	15.56 (25)	7.4 (13)	1.1 (31)	.271

Note. CAPE-V = Consensus Auditory–Perceptual Evaluation of Voice.

Results

Adherence: Practice Frequency and Duration

Over the course of the study, participants in the app group practiced on average twice as often (M = 30.38, SD = 13) as control participants (M = 15.3, SD = 8.9), a significant difference with F(1, 32) = 15.151, p = .000, commensurate to an effect size of 1.35 (Cohen's d). Sequential examination of each week of practice demonstrated significantly greater practice for app participants, as shown in Table 6. Box and whisker plots in Figure 5 illustrate the differences in weekly practice.

Table 6.

Practice frequency by week.

Measure	Control group M (SD)	App group M (SD)	t (df)	p	Cohen's d
Number of practice sessions Week 1	4.17 (3.7)	11.88 (6.5)	−4.16 (32)	.000*	1.457
Number of practice sessions Week 2	6.28 (3.06)	8.44 (3.6)	−1.8 (32)	.017*	0.6465
Number of practice sessions Week 3	4.89 (4.17)	10.06 (6.5)	−2.8 (32)	.002*	0.9113

^* p < .05.

Figure 5.

Weekly group differences in practice frequency.

App use was associated with a slight reduction in practice session duration, with control sessions lasting, on average, 4.89 min (SD = 2.5) per practice session, while app practice sessions were 3.38 min per session (SD = 2). This difference was not significant, F(1, 31) = 3.418, p = .074.

Motivation for Practice

Self-efficacy for practice was comparable for both groups at study onset and slightly higher for both groups at study completion (see Table 7). A repeated-measures ANOVA was conducted to examine the effect of time and homework condition on participant self-efficacy for practice. Box's M of .609, F(3, 363398.517) = 0.189, p = .904, demonstrated homogeneity of variance. Self-efficacy did not increase significantly over time, as Wilks' lambda value was .906, F(1, 32) = 3.309, p = .078. The difference between groups was not significant, F(1, 32) = 0.618, p = .437, and there was no significant interaction between time and condition. Thus, self-efficacy neither grew significantly over time, nor did it differ significantly by group.

Table 7.

Dependent variable means at study onset and completion for both groups.

Measure	Control group M (SD) Pretherapy	App group M (SD) Pretherapy	Control group M (SD) Posttherapy	App group M (SD) Posttherapy
Voice Handicap Index (pretherapy)	36.11 (18.30)	33.19 (10.50)	28.83 (16.7)	21.19 (10.5)
CAPE-V (pretherapy)	11.07 (12.6)	8.67 (7.2)	6.347 (7.09)	5.035 (6.7)
Readiness Ruler Goal Commitment: Practice	8.44 (1.4)	8.48 (1.3)	7.33 (2.0)	7.47 (2.07)*
Readiness Ruler Goal Commitment: Generalization	6.61 (2.5)	6.96 (3.1)	7.78 (2.2)	7.44 (2.2)*
Self-Efficacy Scale: Practice	6.43 (2.1)	6.08 (1.7)	7 (1.7)	6.32 (1.8)
Self-Efficacy Scale: Generalization	4.57 (3.1)	3.89 (3)	7.3 (1.6)	7.1 (1.4)
Self-perceived generalization %	15.56 (25)	7.4 (13)	59% (26)	66% (19)

Note. CAPE-V = Consensus Auditory–Perceptual Evaluation of Voice.

^* p < .05.

Likewise, goal commitment for practice was examined in a similar fashion. Box's M demonstrated homogeneity of variance at 1.715, F(3, 363398.517) = 0.533, p = .66. As noted in Table 7, goal commitment exceeded a mean of 8 for both groups at study onset but lowered slightly by study completion. This reduction of time was significant, with Wilks' lambda of .815, F(1, 32) = 7.261, p = .011. However, the reduction did not differ by group, with F(1, 32) = 0.042, p = .84, and eta squared = .001. The interaction between time and condition was not significant, F(1, 32) = 0.0150, p = .904, nor was there an effect size with and partial eta squared at .000. Thus, goal commitment for practice reduced as participants shifted their focus from practice to implementation in daily communication, but group differences were not observed.

Self-Reported Generalization

As shown in Table 7, participants in the written homework group reported slightly greater use of resonant voice at baseline than the app group. Conversely, app participants reported more use of the target voice at study conclusion—66% (SD = 19%)—than control participants' 59% (SD = 26). Given an unequal baseline, change scores were calculated, yielding a self-reported gain in generalization of 59% (SD = 22) for app participants and 43% gain (SD = 31) for controls.

A repeated-measures ANOVA was used to examine the effect of time and homework condition on self-reported generalization. An assumption of homogeneity is confirmed using Box's M of 7.01, F(3, 363398.517) = 2.179, p = .088. Since the Box's M alpha level is .001, the null hypothesis indicating homogeneity of variance is retained. There was a significant effect of time; Wilks' lambda is .01, F(1, 32) = 127.484, p = .001, indicating that both groups increased significantly over time in self-reported generalization.. The effect size measured using partial eta squared was .799, which is large according to Cohen. There was appropriate power for the test: a value of 1. Figure 6 shows that both groups significantly increased in their use of the target voice, with the app group overtaking the written homework group during the course of the study. However as illustrated by the graph, this between-groups difference upon study completion was nonsignificant at F(1, 32) = 0.026, p = .872. The interaction of time and condition was also not significant with F(1, 32) = 2.77, p = .106, with a medium effect size of .08 indicated by partial eta squared, but underpowered at a power of .365.

Figure 6.

Effect of the app on self-reported generalization

Motivation for Intentional Implementation

Self-efficacy for intentional implementation of resonant voice in daily communication was tested in a similar fashion with Box's M of 1.752, F(3, 363398.517) = 0.544, p = .652. There was a significant effect of time, with Wilk's lambda of .936, F(1, 32) = 36.208, p = .000, indicating that both groups gained a significant amount of self-efficacy for generalization over time. The effect size was large, with partial eta squared = .531, and power appropriate at a value of 1. Self-efficacy did not differ significantly per group, with F(1, 32) = 1.403, p = .245, and there was no interaction between time and condition.

Goal commitment for implementation, Box's M = 1.196, F(3, 363398.517) = 0.371, p = .774, increased significantly over time with Wilk's lambda = .841, F(1, 32) = 5.857, p = .022. The effect size was large, with a partial eta squared of .159. As with the other motivation measures, there was no between-subjects difference with F(1, 32) = 0.193, p = .663, and no interaction between time and condition, F(1, 32) = 0.051, p = .823.

Voice Therapy Outcomes

For examination of VHI score changes, an assumption of homogeneity is confirmed using Box's M of 10.754, F(3, 363398.517) = 3.341, p = .018. We can see that both groups had a reduction of VHI score during the 3-week period in Table 7. VHI for both groups reduced significantly, as evidenced by the overall Wilk’s lambda of .703, F(1, 32) = 13.522, p = .001, indicating benefit from therapy. The effect size measured using partial eta squared was .297, which indicates a large effect size according to Cohen. There was appropriate power for the test (.946). The means for the control and treatment group on the pre-VHI and post-VHI scores are shown in Figure 7.

Figure 7.

Effect of the app on Voice Handicap Index (VHI) score reduction.

Although VHI declined more greatly in the app group, the between-group difference was nonsignificant at F(1, 32) = 1.237, p = .274. However, the partial eta squared was .037, indicating a small effect size, and the test was underpowered (.190). These finding suggest that the sample size would need to be larger to determine if there is a statistically significant reduction between the groups. From this study, we can only determine that there is a slightly greater reduction in VHI scores. As an exploratory measure to guide future research, a power analysis was conducted, revealing that an N of 102 (i.e., 51 participants per group) is needed to detect a medium effect (Cohen's d = 0.5) with 80% power.

For CAPE-V reduction, the overall test showed a decrease in the CAPE-V scores on average for both groups; however, the change in the CAPE-V means from pre to post was nonsignificant, with Wilk's lambda of .908, F(1, 31) = 3.141, p = .086. Though the effect size (ή² = .092) was moderate, the test was underpowered (.404), suggesting that a larger sample was needed to show any statistically significant difference. The between-groups difference was also nonsignificant, F(1, 31) = 0.445, p = .510.

App Usability and Utility: SUS

SUS scores averaged 77.68 (SD = 17.8), corresponding to an adjective rating of “good” (Bangor et al., 2009). Scores ranged substantially from 42.5 (fair) to 100 (best imaginable). In a post hoc exploratory analysis, the association between SUS scores and practice variables was determined through calculation of Spearman rho. A significant, moderate relationship between perceived usability and total number of practice sessions (r_s = .667, p < .01) was found. No relationship was found between SUS scores and duration of practice sessions.

Usability and Utility: Interview Comments

Overall, the app was perceived as helpful, containing both benefits and limitations in utility and usability. Reminder alerts were experienced as useful in providing self-regulatory support even when practice could not logistically be completed at the time of the reminder. In fact, one participant noted that “having a physical reminder” of voice therapy (i.e., borrowing an iPod from the laboratory) was useful in and of itself. Two participants stated that reminders were unnecessary for them, because their practice was typically prevented by logistic rather than memory barriers. All qualitative results are provided in Table 8.

Table 8.

Excerpts of qualitative results from app semistructured interviews.

Theme	Subthemes	Example
Reminders are useful	Self-regulatory support/memory aid Raises consciousness	“If it wouldn't have popped up on my phone, I would have totally forgotten.” “Knowing there's a reminder…(you) don't stress about how I'm going to remember one more thing.” “It reminded me that I'm in voice therapy,…even when I couldn't do it then.” (i.e., at time of reminder)
Reminders unnecessary	Logistics, not forgetting is barrier Practice is demanding	“When I didn't practice it wasn't because I didn't remember: something else got in the way.” “It's easy to do daily but hard to do more than that.” “It's partly mental, partly emotional, and partly physical…it takes a lot out of you.”
Audio instructions and examples are useful	Examples/instructions aid recollection of exercise Recordings provide voice model Exercise set provided	“It helped you remember (content) because as soon as I'd get home I was like ‘what was I supposed to do again?’”… “The difference between practicing after the first session (i.e., patient had therapy elsewhere) and not practicing in the app.” “When I hadn't done an exercise for a few days I had to review the instructions.” “It's prompting the right voice.” “There was a structure I could follow and there was a timer so I'd have a sense of time and you could repeat a recording or add time….”
Instructions not needed	Instructions not needed once exercise is acquired	“I didn't need instruction every time.”
CPP useful	Objective feedback Judge progress	“It provides guidance,” “The number helps you know if I want to get it forward; a brighter sound…the CPP helps to measure it….” “The statistics help you kind of teach yourself….” “I'd go back and redo them….“ (if good score not reached) “I like seeing because you do one in the beginning and one in the end…often at the end you'd do better.”
CPP not useful	Meaning of CPP unclear CPP values invalid Kinesthetic knowledge needed regardless	“I wasn't sure what it measured….” “I don't always trust it. Sometimes I thought a sound was better but the number was lower….” “You need to feel it: you can't just rely on numbers….” “It's not essential because I have to be able to feel how it's buzzing in my ear and in front….” “There's no real-time feedback in the exercises so it's my subjective assessment of whether I'm following instructions while in the session I get constant feedback….”
iPod	Accountability	“It's good to have a device that keeps you accountable….” “It was nice to have a physical thing….”
Technical issues	Flexibility limitations exercise Timer limitations (same times daily) Instruction skip button needed (was resolved later that week)	“I wanted to change the order of exercises. Sometimes you need an exercise, sometimes you don't.” “I have a different schedule every day (i.e., reminders were same time daily; could not adjust).” “To have to listen to the instructions every time (before practice time) is annoying.”

Note. CPP = cepstral peak prominence.

With regard to the utility of recorded practice instructions, patients found these helpful overall and particularly at the start of therapy (i.e., when voice exercises were new) or when they had not practiced for some time and forgotten details of the exercise. Therefore, patients found utility in the “skip” button for instructions, as instructions were not needed consistently.

Usefulness of the CPP voice test was reported by three participants for whom CPP values differentiated their habitual voice quality from their target production or for whom values improved (i.e., increased in value) between the start and completion of a practice session. As one participant stated, the CPP value provided an objective “gauge” for self-evaluation of voice technique. However, participants also noted that such external feedback was not adequate to ensure vocal improvement; rather, how the voice “feels” (i.e., kinesthetic feedback) was ultimately required for successful voice use. CPP was not deemed useful by the majority of participants, who noted, for example, “I didn't really know what the numbers meant,” indicating confusion.

Suggestions for improvement were focused on flexibility of functionalities. Due to technical limitations, reminder alerts sounded at the same time each day and could not vary by daily schedule. Consequently, reminders could occur at inopportune times, reducing their usefulness. Also, the order of exercises was set by the therapist. Although participants could skip instructions and exercises and both reduce or increase practice durations for each exercise at the time of practice, they expressed desire for increased autonomy. Several participants requested additional acoustic parameters. Another requested that CPP would be calculated for all practice recordings, not only those completed within the CPP functionality. Lastly, participants reported instances in which the app failed secondary to bugs or iOS upgrades.

Discussion

Mobile apps are increasingly utilized in health behavior interventions, but less commonly investigated for their efficacy (Decker & Buggey, 2014). An app was designed to support homework adherence via reminders, recorded examples, and CPP feedback. This study examines the effect of this app on adherence behavior and associated motivation as compared to written homework instructions, as well as its usability and utility. Results indicated that the app significantly increased practice frequency but did so in the absence of fostering increased patient motivation. Overall usability and utility were good, with SUS scores significantly and moderately predictive of practice frequency.

Effect on Practice

The app's effect on practice was substantive, as indicated by large effect sizes for weekly practice. These effects exceed the small to medium effect sizes reported for health behavior apps in meta-analyses and reviews in the wider health behavior literature (Badawy & Kuhns, 2017; Fanning et al., 2012; Fedele et al., 2017; Mateo et al., 2015; Miller et al., 2017; Nicholas et al., 2015). The voice app's greater effect sizes may be attributed to the close relationship between its features and patients' needs. Features addressed known barriers to voice therapy adherence (van Leer & Connor, 2010), were evidence-based in voice therapy (van Leer & Connor, 2012, 2015; van Leer et al., 2017) or in the larger field of behavior change (Kannisto et al., 2014; Pop-Eleches et al., 2011; Vervloet, Linn, et al., 2012; Vervloet, van Dijk, et al., 2012), and grounded in social cognitive theory. The app's development thus represents an iterative, patient-centered process that is fundamental to persuasive and human-centered design (Fogg, 2002; Harte et al., 2017), in particular when the goal is to trigger a target behavior (Fogg, 2002). The moderate, significant association between SUS scores and practice frequency confirms the importance of the app's usability and utility in eliciting practice behavior.

Comparison to MP4 Support

Results of this study can be readily compared to those of the MP4 study (van Leer & Connor, 2015) because study designs were identical. App support far exceeded MP4 support in its effect on practice frequency but, unlike MP4 support, did not significantly affect self-reported generalization or any aspect of motivation. One would expect greater self-efficacy for practice associated with significantly more frequent practice in this study. In order to explain these similarities and differences in findings, the content of the MP4 and app support systems must be compared.

As potential key ingredients to influence adherence and motivation, the app provided (a) recorded examples (i.e., instructional recordings), (b) reminders notifications, and (c) CPP feedback. However, qualitative poststudy interview data revealed that the majority of study participants did not find CPP feedback comparing their habitual and resonant voice useful (i.e., “I didn't really understand what the numbers meant”). As such, CPP is unlikely to have played a substantial role in this study. This leaves recorded examples and reminder notifications as the two key ingredients.

Compared to the MP4 intervention, which was composed only of example recordings, the app's additional key ingredient was its reminder notifications. It logically follows that this addition resulted in the app's greater (and large) effect on practice. Results can be compared to those of a physical therapy adherence study, in which the combination of reminders and photo examples of exercises yielded significantly more frequent practice than the standard-of-care provision of photo handouts alone (Lambert et al., 2017). It is not surprising that adding reminders strengthened the intervention, because reminders alone have demonstrated efficacy in triggering a wide variety of behaviors by providing external self-regulatory support (Kannisto et al., 2014; Pop-Eleches et al., 2011; Vervloet, van Dijk, et al., 2012). When combined with instructional examples, important barriers to practice were addressed.

Lack of Effect on Motivation for Practice

Given the doubling of practice, the absence of an increase in motivation was unexpected. Findings were not consistent with positive effects on motivation found in both MP4 studies (van Leer & Connor, 2012, 2015), the CPP study (van Leer et al., 2017), and social cognitive theory's premise that successful behavior yields increased self-efficacy (Bandura, 2001). There are several possible explanations. First, although app participants practiced twice as often as controls, they nonetheless practiced half as often as recommended. Thrice daily reminder notifications may have emphasized this failure, thus restricting the expected growth of self-efficacy. Second, compared to the MP4 intervention, the app's use of audio instead of video models may have reduced the motivational impact of instructional and model recordings. While video MP4 examples were described as “motivating” and “encouraging” by MP4 trial participants (van Leer & Connor, 2012, 2015), audio recordings received no such praise in this study. The positive effect of video examples on self-efficacy is well established across behavioral domains (Bellini & Akullian, 2007), in particular for the self-as-model (Prater et al., 2012), but the motivational effect of audio models in unknown. Furthermore, peer model interviews present in the MP4 trial were not included in the app. Peers discussed a variety of challenges and ways to overcome or manage these, including difficulties with practice and implementation, and maintaining motivation. This content may have served as a powerful motivating ingredient. More research is needed to understand the role of various video versus audio examples in voice adherence tools.

Self-Reported Generalization and Associated Motivation

Although patients in the app group perceived themselves as generalizing resonant voice more than those in the written homework group, the difference was not significant in this study, while it was in the MP4 study. Again, video format and peer support may have been needed to affect implementation of resonant voice throughout the day or affect perceptions of implementation. Since patient perceptions of progress relate to adherence and drop out, these findings are directly relevant to clinical practice (Hapner et al., 2009). However, they should be interpreted with caution, given the self-report nature of measurement. How well patient perceptions corresponded with actual generalization requires validation, for example, through comparison with samples of daily voice use obtained covertly via electronically activated recording (Mehl et al., 2001). App-based simulated phone calls can also sample the voice throughout the day, as well as support implementation directly (van Leer & Porcaro, 2019). Furthermore, ambulatory feedback approaches hold great promise for both tracking and influencing the daily use of a target voice technique by assessing hyperfunction via inverse filtering of ambulatory accelerometer signals and providing summary feedback to the wearer (Ghassemi et al., 2014).

Effects of the App Within the Triadic Model

Since only a behavioral target was reached, the app's mechanism of action can be considered entirely behavioral in nature, rather than both behavioral and motivational as intended. Qualitative data confirm that “the app was like a nagging spouse,” triggering task completion without instilling positive beliefs about these tasks. Figure 8 illustrates this behavioral mechanism of action within the framework of social cognitive theory's model of triadic reciprocal causation.

Figure 8.

Behavioral effect of the app in the triadic model.

Role of CPP

Qualitative study results showed that most participants had difficulty interpreting CPP results. Confusion may have resulted from several sources. Since CPP feedback for resonant voice was interpreted relative to values for each participant's habitual voice, participants may have needed both more familiarity with producing and differentiating these modes of phonation for CPP to be meaningfully incorporated into their practice. Indeed, participants in the CPP feasibility study—who found CPP encouraging and useful—did have this experiences, as they had completed both a minimum of two therapy sessions and had reached the generalization phase of therapy at time of study participation (van Leer et al., 2017).

Furthermore, app participants' mild dysphonia level may not have yielded easily discernable differences between the habitual and target voice, in particular, given possible variability in mouth-to-mic distance and acoustic environments. To detect subtle differences measures other than CPP may also have potential utility. For example, pitch strength (Shrivastav & Sapienza, 2006) and relative fundamental frequency at voicing on and offset (Stepp et al., 2011) might be more sensitive to subtle differences in mildly aperiodic voices. For near-normal voices, direct measurement of the energy in the speaker's formant area associated with resonant voice (Barrichelo & Behlau, 2007; Smith et al., 2005) might have been useful. However, given that resonant targets are not as extreme in voice therapy as in the Lessac “call” (Raphael & Scherer, 1987) and Lessac “Y-buzz” (Barrichelo-Lindström & Behlau, 2009), these approaches may have failed as well. In summary, study results regarding CPP utility and usability suggest inclusion of additional measurement approaches and increased training time in the future.

Clinical Outcomes

It is premature to draw any conclusions regarding effects on clinical outcomes from this initial study. Since the purpose of the study was to examine the effect of homework modality on adherence, the study was powered to detect these effects. A power analysis revealed that more subjects (51 per group) are needed to detect medium effects in VHI score reductions. Therefore, a larger N is needed in future research. It is certainly interesting that twice the practice did not yield significant group differences in VHI scores. Although it is possible that the dose difference, while large, was not great enough to yield downstream effects on outcome measures, it is also possible that a full course of therapy is needed to detect adequate effects. Extending the study period to six to eight sessions may be necessary for self-reported generalization to occur. Thus, a next step in seeking outcome differences is to increase app features and study duration, as well as studying a larger sample size.

Clinical Implications

Results of this study do hold direct implications for clinical practice. Specifically, the positive effect of the app on practice frequency supports its provision to patients who struggle to complete daily practice. The app per se is not needed: Calendars and digital recording functionalities are available on all smartphones. Using these in combination is highly likely to assist patients in completing practice, as it did in this study. Comparison of study results with previous research suggests that video models, including peer interviews, may be needed to enhance motivation and perceived generalization. These again can be created with standard smartphone apps. Feedback functionalities such as CPP may be useful but are likely to require more extensive training and evaluation of appropriateness. Mobile support may also hold value for postdischarge maintenance of target voice technique.

Further Insights Into Adherence

In addition to the planned research questions, study results shed light on voice therapy adherence in general. Analysis of homework recordings revealed substantially less practice than was documented via self-report in our previous studies (van Leer & Connor, 2012, 2015). Across areas of adherence research, self-report is known to represent an exaggeration of actual adherence values (Kazantzis et al., 2004). Several of our control participants misrepresented the timing of their recorded practice sessions, for example, by labeling these as daily (e.g., “Monday practice, Tuesday practice”), while time–date stamps revealed that these were completed within 1 hr prior to the therapy session. Thus, time–date stamps covertly tracked actual practice times at odds with self-reported times. Misrepresentation of an activity schedule has also been demonstrated in a study comparing diary entry dates to covertly measured diary opening times in an electronically wired diary (Stone et al., 2003). Since voice practice is known to increase when patients must submit recordings of their practice (Ellis & Beltyukova, 2011), practice frequency of clinical voice therapy patients may be even less than those observed in a study, if the patients do not need to submit recorded evidence of practice.

Limitations and Future Directions

Study results are focused in several ways. Results are relevant to individuals undergoing a short course of direct resonant voice therapy and may alter with longer study durations or other treatment approaches. Future study is needed to investigate patients' need for mobile support over a longer course of therapy. Results are also limited to individuals with mild dysphonia. Individuals with greater dysphonia may derive more utility from CPP feedback. Generalized use of the target voice was measured only through a patient-perceived estimation identical to our previous studies. Lastly, practice accuracy was not evaluated in this study. An approach is under development to assess practice accuracy of practice recordings obtained in divergent acoustic environments.

Conclusions

Strong evidence was provided for the use of an iOS app to practice a resonant voice protocol. Quantitative and qualitative results suggest that the key effective ingredients of the app were its reminder alerts and recorded instructions, while benefit from CPP may be restricted to particular patient scenarios. The absence of effects on motivation may suggest the utility of using video instead of or in addition to audio recordings for provision of models and instructions. Given the ease of providing comparable support with existing commercial apps, feasibility is excellent, and implications for clinical practice are substantial.

Author Contributions

Eva van Leer: Conceptualization (Lead), Data curation and Formal Analysis (Lead, shared with Britney Lewis and statistical consultant Robert Clayton Hendrick), Funding acquisition (Lead), Investigation (Lead), Methodology (Lead), Project administration (Lead), Resources (Lead) Supervision (Lead), Visualization (shared with Brittney Lewis), Writing – original draft (Lead) Writing – review & editing (Lead, shared with Brittney Lewis). Britney Lewis: Data curation (Supporting), Formal analysis (Supporting), Writing – review & editing (Supporting). Nick Porcaro: Software (Supporting), Writing – original draft (Supporting), Writing – review & editing (Supporting).

Funding Statement

This study was funded by National Institute on Deafness and Other Communication Disorders Grant 1 RO3 DC013884-01 (awarded to PI: Eva van Leer).