Validation of an AI-assisted Treatment Outcome Measure for Gender-Affirming Voice Care: Comparing AI Accuracy to Listener's Perception of Voice Femininity.

OBJECTIVES: There is currently a lack of objective treatment outcome measures for transgender individuals undergoing gender-affirming voice care. Recently, Bensoussan et al developed an AI model that is able to generate a voice femininity rating based on a short voice sample provided through a smartphone application. The purpose of this study was to examine the feasibility of using this model as a treatment outcome measure by comparing its performance to human listeners. Additionally, we examined the effect of two different training datasets on the model's accuracy and performance when presented with external data. METHODS: 100 voice recordings from 50 cisgender males and 50 cisgender females were retrospectively collected from patients presenting at a university voice clinic for reasons other than dysphonia. The recordings were evaluated by expert and naïve human listeners, who rated each voice based on how sure they were the voice belonged to a female speaker (% voice femininity [R]). Human ratings were compared to ratings generated by (1) the AI model trained on a high-quality low-quantity dataset (voices from the Perceptual Voice Quality Database) (PVQD model), and (2) the AI model trained on a low-quality high-quantity dataset (voices from the Mozilla Common Voice database) (Mozilla model). Ambiguity scores were calculated as the absolute value of the difference between the rating and certainty (0 or 100%). RESULTS: Both expert and naïve listeners achieved 100% accuracy in identifying voice gender based on a binary classification (female >50% voice femininity [R]). In comparison, the Mozilla-trained model achieved 92% accuracy and the previously published PVQD model achieved 84% accuracy in determining voice gender (female >50% AI voice femininity). While both AI models correlated with human ratings, the Mozilla-trained model showed a stronger correlation as well as lower overall rating ambiguity than the PVQD-trained model. The Mozilla model also appeared to handle pitch information in a similar way to human raters. CONCLUSIONS: The AI model predicted voice gender with high accuracy when compared to human listeners and has potential as a useful outcome measure for transgender individuals receiving gender-affirming voice training. The Mozilla-trained model performed better than the PVQD-trained model, indicating that for binary classification tasks, the quantity of data may influence accuracy more than the quality of the data used for training the voice AI models.

Perceptions of Vocal Performance Impairment in Singers with and without Hearing Loss.

INTRODUCTION: It is well understood that hearing plays an important role in accurate vocal production. Singers in particular rely on auditory cues and auditory feedback to reproduce specific pitch contours. Therefore, even mild hearing loss may have a detrimental effect on a singer's ability to perform. This study investigates the effect of hearing loss on self-reported vocal production in singers, specifically in the domains of pitch matching, pitch maintenance, vowel production, and dynamic control as well as the effect of hearing loss on vocal handicap. METHODS: An 18-item electronic self-report survey was distributed to the members of the National Association of Teachers of Singing and to the Voice Foundation. Data collected included demographics, the Singing Voice Handicap Index-10 (SVHI-10), and a series of closed and open-ended questions. Demographic variables, variables related to the SVHI-10, and variables related to the newly introduced survey were included in a hierarchical regression analysis to determine significant relationships. RESULTS: Among 206 eligible participants, 37 individuals reported a voice problem, 58 reported hearing loss, and 19 reported concurrent hearing loss and a voice problem. Among males, there were no significant differences between hearing impaired and normal hearing singers in reported pitch matching, pitch maintenance, dynamic control, and vowel matching when those with voice problems were excluded and included. However, in females, when singers with voice problems were excluded, there was a significant difference between hearing impaired and normal hearing singers in pitch matching (P = 0.38). Additionally, when singers with voice problems were included in the female subset, significant differences emerged between the hearing impaired and normal hearing singers in areas of pitch matching (P = 0.01) and vowel matching (P = 0.02). Further, controlling for gender, when excluding voice problems, there was a significant difference between the SVHI-10 scores of normal hearing (mean = 9.03) and hearing impaired participants (mean = 11.30, P = 0.02). This difference continued to be significant when including those with voice problems (normal hearing mean = 9.97, hearing impaired mean=14.1, P <0.0001). Additionally, individuals with hearing impairments were more likely to report higher perceived vocal handicap scores as reflected on the SVHI-10 than normal hearing respondents (P = 0.002). Other factors associated with higher likelihood of SVHI-10 score include older age (P = 0.008), having a voice problem (P <0.0001), and being paid to sing within the past six months (P = 0.001). CONCLUSION: When controlling for voice problems, singers with hearing impairments subjectively did not perceive that they performed less accurately on pitch matching, pitch maintenance, dynamic control, and vowel matching, yet they scored higher on the SVHI-10 indicating vocal handicap. Further study is needed to characterize the relationship between perceived and measured vocal accuracy in singers with hearing loss.

Gauging Stakeholder Perspectives on Artificial Intelligence in Gender-Affirming Voice Care.

BACKGROUND: As the main objective outcome measure used in gender-affirming voice care (GAVC), fundamental frequency (f0 ) often fails to accurately reflect patient perceptions of their voice. Our team developed an artificial intelligence (AI) program that provides an alternative objective outcome measure that has the potential to more accurately align with patient perceptions. OBJECTIVE: To gauge stakeholder receptivity to the use of AI in GAVC before employing a novel outcome measure in transgender and nonbinary communities. METHODS: This prospective qualitative study used online focus groups composed of speech-language pathologists (SLPs), transgender men (TGM), transgender women (TGW), and nonbinary (NB) individuals. Participant age, race, gender, and geographic location were recorded. Each cohort participated in a series of two focus group sessions. The first session focused on participant experiences in GAVC, whereas the second ascertained participant perspectives on the use of AI in GAVC. Transcripts of each discussion were coded using Nvivo to perform inductive thematic analysis. RESULTS: Seven SLPs, seven TGW, three TGM, and two NB individuals (mean [range] age, 35.5 [26-48] years) participated. Transgender and nonbinary participants were generally amenable to the technology, whereas SLPs were more hesitant about its use. Positive findings included appreciation for AI as an objective outcome measure and enthusiasm for its potential to longitudinally track progress. Hesitations concerned the actionability of using the AI and unease about the black box nature of the AI's analysis. CONCLUSION: Transgender and NB individuals were receptive to the use of AI technology in GAVC, whereas SLPs were more apprehensive about using AI. LEVEL OF EVIDENCE: NA Laryngoscope, 133:1698-1705, 2023.

Development of Cushing Syndrome After Serial In-office Subglottic Steroid Injections.

In-office subglottic intralesional steroid injections (SILSI) have gained popularity as an adjunct to operating room dilation in the treatment of subglottic stenosis. They are generally thought to have a low risk profile for development of systemic side effects. Here, we present a case of a 55 year old woman who developed symptoms of Cushing syndrome after receiving SILSI, including weight gain, striae, dorsal hump and alopecia. This case illustrates that despite the localized nature of SILSI, there is still a risk of developing systemic effects as a result of the treatment. Laryngoscope, 132:942-943, 2022.

Regression of intestinal metaplasia following magnetic sphincter augmentation device placement.

BACKGROUND: Intestinal metaplasia represents an esophageal mucosal transformation due to uncontrolled gastroesophageal reflux disease. Fundoplication has been shown to lead to regression of disease. Magnetic sphincter augmentation is an alternative to fundoplication that effectively treats reflux disease. Initially, patients with intestinal metaplasia were not considered candidates for device placement, so outcomes in these patients are unknown. METHODS: A retrospective review of all patients who underwent magnetic sphincter augmentation device placement between 2007 and 2017 was performed. All patients underwent pre-operative endoscopic evaluation and were categorized as having ultra-short segment (less than 1 cm), short-segment (1-3 cm), or long-segment (greater than or equal to 3 cm) disease. To be included in the study, pathologic examination demonstrating columnar mucosa with goblet cells was required. RESULTS: There were 86 patients with biopsy-proven non-dysplastic intestinal metaplasia. 35 patients had ultra-short segment, 37 patients had short-segment, and 14 patients had long-segment disease. At a median follow-up of 1.2 years, 67/86 (78%) patients completed endoscopic follow-up. 48/67 (71.6%) patients had regression of intestinal metaplasia. There was no progression to dysplasia or carcinoma. Patients with abnormal post-operative DeMeester scores were less likely to have regression of disease. Regression was more likely in the ultra-short segment (82.8%) and short-segment (73.3%) groups compared to the long-segment group (25.0%). CONCLUSIONS: Magnetic sphincter augmentation is effective in achieving regression of intestinal metaplasia. Longer-term follow-up is needed to assess durability of effect and make meaningful comparisons to fundoplication.