Umeå University's logo

Purpose: This study aimed to determine the acoustic properties most indicative of dysprosody severity in patients with Parkinson's disease using an automated acoustic assessment procedure.

Method: A total of 108 read speech recordings of 68 speakers with PD (45 male, 23 female, aged 65.0 ± 9.8 years) were made with active levodopa treatment. A total of 40 of the patients were additionally recorded without levodopa treatment to increase the range of dysprosody severity in the sample. Four human clinical experts independently assessed the patients' recordings in terms of dysprosody severity. Separately, a speech processing pipeline extracted the acoustic properties of prosodic relevance from automatically identified portions of speech used as utterance proxies. Five machine learning models were trained on 75% of speech portions and the perceptual evaluations of the speaker's dysprosody severity in a 10-fold cross-validation procedure. They were evaluated regarding their ability to predict the perceptual assessments of recordings excluded during training. The models' performances were assessed by their ability to accurately predict clinical experts' dysprosody severity assessments.

Results: The acoustic predictors of importance spanned several acoustic domains of prosodic relevance, with the variability in fo change between intonational turning points and the average first Mel-frequency cepstral coefficient at these points being the two top predictors. While predominant in the literature, variability in utterance-wide fo was found to be only the fifth strongest predictor.

Conclusion: Human expert raters' assessments of dysprosody can be approximated by the automated procedure, affording application in clinical settings where an experienced expert is unavailable. Variability in pitch does not adequately describe the level of dysprosody due to Parkinson's disease.

Gender-affirming voice training aims to assist transgender and gender-diverse clients in reaching a voice in better alignment with their gender identity. We describe the development and evaluation of the Auditory-Perceptual Assessment of Gender Expression in Voice (PAGE-V) protocol for assessing voice and speech features commonly targeted in gender-affirming voice training. Features previously identified to influence the perception of femininity and masculinity were evaluated in two steps. First, seven speech and language pathologists (SLPs) reviewed PAGE-V based on their experience in providing gender-affirming voice training. Second, six SLPs completed listener training and assessed 45 voice samples with varying gender expressions using PAGE-V. The results showed that individual SLPs rated items relatively consistently; agreement among SLPs was lower. The PAGE-V was concluded to provide a clinically useful tool, and the insights gained support further development of PAGE-V. Listener training sessions are needed to strengthen SLPs’ internal representations and consensus on protocol items.

Context: There is a considerable diagnostic delay in acromegaly, contributing to increased morbidity. Voice changes due to orofacial and laryngeal changes are common in acromegaly.

Objective: Our aim was to explore the use of digital voice analysis as a biomarker for acromegaly using broad acoustic analysis and machine learning.

Methods: Voice recordings from patients with acromegaly and matched controls were collected using a mobile phone at Swedish university hospitals. Anthropometric and clinical data and the Voice Handicap Index (VHI) were assessed. Digital voice analysis of a sustained and stable vowel [a] resulted in 3274 parameters, which were used for training of machine learning models classifying the speaker as “acromegaly” or “control.” The machine learning models were trained with 76% of the data and the remaining 24% was used to assess their performance. For comparison, voice recordings of 50 pairs of participants were assessed by 12 experienced endocrinologists.

Results: We included 151 Swedish patients with acromegaly (13% biochemically active and 10% newly diagnosed) and 139 matched controls. The machine learning model identified patients with acromegaly more accurately (area under the receiver operating curve [ROC AUC] 0.84) than experienced endocrinologists (ROC AUC 0.69). Self-reported voice problems were more pronounced in patients with acromegaly than matched controls (median VHI 6 vs 2, P < .01) with higher prevalence of clinically significant voice handicap (VHI ≥20: 22.5% vs 3.6%).

Conclusion: Digital voice analysis can identify patients with acromegaly from short voice recordings with high accuracy. Patients with acromegaly experience more voice disorders than matched controls.

We aimed to fine-tuning the Timed Water Swallow Test (TWST) screening procedure to provide the most reliable prediction of the Flexible Endoscopic Evaluation of Swallowing (FEES) assessment outcomes, with age, sex, and the presence of clinical signs of dysphagia being considered in the assessment. Participants were healthy people and patients with suspected dysphagia. TWST performance and participants' reported dysphagia symptoms were assessed in terms of their utility in predicting the outcome of a FEES assessment the same day. The FEES assessors were blinded to the nature of the TWST performance. The water swallowing capacity levels and clinical observations during a screening performance that were indicative of dysphagia/no symptoms in FEES were determined. Convergent validity was assessed as the agreement with the Functional Oral Intake Scale (FOIS) in the FEES assessment. TWST predicted FEES findings (aspiration and dysphagia) with a sensitivity of 72 and 45% and a specificity of 75% and 80%, respectively. Extended analysis of the TWST procedure (eTWST) identified aspiration (sensitivity = 92%, specificity = 62%) and dysphagia (sensitivity = 70%, and specificity = 72%) more accurately and showed a high correlation with FOIS (ɸ = 0.37). Excellent inter-rater reliability was further observed (Kw = 0.83). The extended evaluation of TWST performance has superior criterion validity to that of TWST. eTWST displayed high convergent validity and excellent interrater reliability. We therefore believe that eTWST can be highly relevant for clinical dysphagia screening.

The Acoustic Voice Quality Index (AVQI) is a widely adopted tool for assessing dysphonia, incorporating sustained vowel and continuous speech samples to enhance ecological validity. Despite its broad use, the reliability of AVQI measurements, particularly in digital health applications, remains underexplored. This study aims to review the literature on AVQI's development and validation from the perspective of internal consistency of acoustic measurements, and to assess the amount of speech required to reliably determine the AVQI of a voice recording. Two sub-studies are described. Study 1: A narrative review was conducted using Google Scholar and Scopus to identify studies mentioning “AVQI” and “reliability”. Key methodological details were extracted from studies reporting AVQI measurements, summarized, and discussed in terms of how the internal consistency of acoustic measurements was ensured across studies where AVQI had been validated or applied. Study 2: Recordings of read and spontaneous speech as well as sustained vowels produced by 54 native Swedish speakers (22 female, 32 male; age range: 46–78 years) were assessed in terms of the amount of speech required to obtain a reliable acoustic measurement of the speakers' voices. Simulations were performed using read and spontaneous speech materials of varying lengths. The variability in AVQI and its sub-measures was analyzed relative to the length (in words or in seconds) of continuous speech included. The result of study 1 shows that out of 129 identified studies, 85 reported on AVQI measurements. The review highlighted substantial variability in continuous speech lengths used across studies, ranging from 4–200 words. A recommendation of 3 s of voiced segments was often adhered to, but was found to lack sufficiently robust grounding. The simulations indicated that AVQI achieves internal consistency at speech lengths of approximately 50 words (or 20 s), which is longer than the current recommendation. Both read and spontaneous speech provided stable AVQI measurements at these speech lengths. AVQI thresholds obtained using speech lengths shorter than 50 words (20 s) may require re-evaluation. Robust standardization of continuous speech lengths is essential for the successful adoption of AVQI in digital health applications.

Visible Speech (VISP) is a web-based research infrastructure hosted at Humlab, Umeå University, designed to enable secure and efficient research on audio recordings of speech. VISP addresses the dual challenge of facilitating advanced linguistic and interdisciplinary research while ensuring compliance with GDPR and Swedish legal requirements for handling personally identifiable and sensitive data. The platform provides a unified environment for storage, controlled access, and reproducible signal processing of speech data, offering one of the most comprehensive sets of speech and voice analysis tools globally. VISP integrates with the Swedish Academic Identity Federation (SWAMID), enabling federated login for researchers across Sweden and supporting collaborative workflows without compromising data security. This allows the processing and analysis of sensitive recordings—such as those involving ethnicity, health, or union membership—within a secure digital framework. Beyond analysis, VISP promotes long-term data stewardship by implementing standardized directory structures and supporting FAIR principles, thereby lowering barriers to archiving and data sharing. As part of the national research infrastructure, Språkbanken and SweCLARIN, VISP contributes to the European CLARIN ERIC ecosystem, strengthening Sweden’s position in digital and experimental humanities. By combining a dedicated software framework, external security reviews, and national identity federation, VISP enables researchers to explore spoken language data in ways previously unattainable, fostering innovation and collaboration across disciplines. 

Spoken language is central to many human interactions and provides the medium through which activities and events across many humanities and social sciences disciplines are studied. It is also the object of active study in itself. As central to humanity as spoken language is, regulations aimed at mitigating privacy concerns also affect the affordance for collaborations on a national or larger scale based on spoken materials.  

The Visible Speech (VISP) platform is a web-based research infrastructure at Humlab, Umeå University, designed to handle audio recordings of speech in compliance with the national implementation of GDPR and security requirements. VISP provides a centralized environment for research of all disciplines in which recordings of spoken language constitute the primary material, meeting both researchers’ needs for efficient workflows and legislators’ demands for secure data management.

One of VISP's primary advantages is its ability to facilitate research on audio recordings that now constitute personally identifiable information (PII) under the application of the GDPR in Sweden. These recordings may contain sensitive content or have been made in sensitive contexts, classifying them as sensitive PII under national legislation. Sensitive contents may occur in relation to, for instance, the ethnicity and religious beliefs of the speaker, and sensitive contexts may occur when the recording is made in a healthcare context or in a context where a person’s membership with a union organization is divulged. While the challenges in conducting larger research efforts on the types of materials are currently aggravated by the implementation of the GDPR locally in Sweden, it is currently not clear to what extent upcoming AI regulation will, in effect, migrate identical or similar constraints to research in other countries in the EU as well.

The VISP platform offers a unified environment for storage, controlled access, direct work, and reproducible speech signal processing. VISP is built on the foundation of earlier research efforts1,2 and includes a comprehensive set of speech and voice analysis procedures within one framework. Thus, national research groups can collectively store interviews or other spoken language recordings, have automatic transcriptions or other speech processing performed, and access the results for complementary manual annotation or analysis simultaneously and securely. Additionally, VISP facilitates the digital archiving of projects through a uniform, documented, and transparent directory structure, reducing barriers to making data available following the FAIR principles. Research projects dealing with sensitive personal data in audio recording form require review by the Ethical Approval Authority and may subsequently take advantage of the VISP facilities.

A significant feature of VISP is its integration with the Swedish Academic Identity Federation (SWAMID, connected to eduGAIN), which enables researchers across Sweden to have secure, federated logins. This national federated login system allows researchers to access project data and collaborate on material processing in ways that were previously not possible. Moreover, VISP supports projects by lowering the step in to digital signal processing and audio analysis of the collected audio signals. This capability allows researchers to perform hands-on processing and analysis without the risk of disseminating sensitive audio recordings. By leveraging SWAMID, VISP ensures that researchers can work seamlessly and securely on collected materials, enhancing collaborative efforts and data handling efficiency. By providing tools for direct manipulation and examination of audio data, VISP ensures that all stages of data handling, from collection to analysis, are conducted within a secure environment, thereby maintaining the integrity and confidentiality of sensitive information.

The work conducted within VISP is part of SweCLARIN, the Swedish node of the European Research Infrastructure Consortium (ERIC) CLARIN. SweCLARIN aims to develop and provide national and European infrastructure for speech and text-based e-science, offering extensive digitized materials and advanced language technology tools. By combining the advanced technologies developed by CLARIN ERIC partners1 with stringent security protocols and leveraging federated login systems, VISP enables efficient and secure research on audio recordings of speech. The VISP components are available for download and setup of local instances and for modification, and the framework, therefore, promises to provide an invaluable tool for researchers, facilitating unprecedented collaboration and data processing within the digital humanities on both a national and larger scale.

References:

[1] R. Winkelmann, J. Harrington, K. Jänsch, EMU-SDMS: Advanced speech database management and analysis in R, Comput. Speech Lang. 45 (2017) 392–410.

[2] R. Winkelmann, J. Harrington, EMU-SDMS: R Centric Semi-automatic Speech Database Processing and Analysis. In Sasha Calhoun, Paola Escudero, Marija Tabain & Paul Warren (eds.) Proceedings of the 19th International Congress of Phonetic Sciences, Melbourne, Australia 2019 (pp. 1317--1321).  Canberra, Australia: Australasian Speech Science and Technology Association Inc

The perceived level of femininity and masculinity is a prominent property by which a speaker’s voice is indexed, and a vocal expression incongruent with the speaker’s gender identity can greatly contribute to gender dysphoria. Our understanding of the acoustic cues to the levels of masculinity and femininity perceived by listeners in voices is not well developed, and an increased understanding of them would benefit communication of therapy goals and evaluation in gender-affirming voice training. We developed a voice bank with 132 voices with a range of levels of femininity and masculinity expressed in the voice, as rated by 121 listeners in independent, individually randomized perceptual evaluations. Acoustic models were developed from measures identified as markers of femininity or masculinity in the literature using penalized regression and tenfold cross-validation procedures. The 223 most important acoustic cues explained 89% and 87% of the variance in the perceived level of femininity and masculinity in the evaluation set, respectively. The median fo was confirmed to provide the primary cue, but other acoustic properties must be considered in accurate models of femininity and masculinity perception. The developed models are proposed to afford communication and evaluation of gender-affirming voice training goals and improve voice synthesis efforts.

Objective: To explore whether cisgender naive listeners, transgender and gender diverse (TGD) listeners, and speech-language pathologists (SLPs) experienced in providing gender-affirming voice training differ in their perception of femininity and masculinity in voices.

Methods: Samples of spontaneous speech were collected from 95 cisgender, and 37 TGD speakers. Three listener groups of cisgender naive (N = 77), TGD (N = 30), and SLP (N = 14) listeners, respectively, rated the voices on visual analog scales in two randomly ordered blocks, in which the perceived degree of femininity was rated separately from the perceived degree of masculinity.

Results: The three listener groups showed similar patterns in their distribution of ratings on the femininity and masculinity scales. The TGD listeners’ mean ratings did not differ from the cisgender naive listeners’, whereas SLPs showed a small, but significant, difference in their ratings compared with both TGD and cisgender naive listeners and rated the voices lower on both the femininity and masculinity scales.

Conclusion: The results differ from previous studies as TGD, and cisgender naive listeners rated the voices very similarly. The lower ratings of femininity and masculinity by the SLPs were likely influenced by their awareness of the complexity in the perception of voices. Therefore, SLPs providing gender-affirming voice training should be attentive to how their professional training may influence their perception of femininity and masculinity in voices and encourage discussions and explorations of the TGD voice client's perceptions of voices.

Background: Without evidence, positive expiratory pressure therapy is a part of rehabilitation worldwide to prevent postoperative hypoxia. Reading aloud could be used as an alternative therapy as lung volumes increases while speaking. We aimed to investigate whether reading aloud is superior to positive expiratory pressure therapy for improving oxygen saturation after abdominal surgery.

Material and Methods: This crossover randomized controlled trial compared reading a text aloud with positive expiratory pressure therapy in patients on postoperative day 1 or 2 after upper gastrointestinal, colorectal, urological, or gynecological abdominal surgery at Umeå University Hospital, Sweden. The primary outcome was the change in peripheral oxygen saturation compared with baseline at 7 min after the intervention. The secondary outcome was transcutaneous carbon dioxide partial pressure change.

Results: This study included 50 patients of which 48 patients were analyzed. Peripheral oxygen saturation rapidly decreased to minimum values below baseline immediately after both interventions and then increased to values above baseline after reading aloud (1%, 95% confidence interval 0.2% to 1%, P = 0.004), but not after positive expiratory pressure therapy (−0.2%, 95% confidence interval −1% to 0.4%, P = 0.436). The difference in oxygen saturation was 1% (95% confidence interval 0.1% to 2%, P = 0.039) at 7 min after termination of the interventions. The interventions reduced transcutaneous carbon dioxide partial pressure by similar amounts.

Conclusions: This trial adds to the evidence against the use of positive expiratory pressure therapy after abdominal surgery. It is even slightly better to read aloud.