interviews · Frontiers in digital health · la publicación, 15 sep 2026 · gratis
“No puedo cambiar la voz porque entonces no sirve”: lo que once empresas de salud por voz contaron sobre los datos que recogen
Investigadores entrevistaron a representantes de once empresas que analizan la voz con inteligencia artificial. Dos no habían pedido revisión ética, cuatro recogían datos en países de ingreso medio y bajo, y el estudio no puede decir qué tan común es nada de esto.
Versión breve · la versión detallada sigue, unos 6 min
Pregunte al chat de weeklyAI
Pregúnteme sobre la IA en su vida: le señalo las notas que weeklyAI ha publicado sobre el tema, esta semana y antes.
Las conversaciones se guardan mientras exista weeklyAI, para mejorar la publicación. Se responde en el idioma en que usted escribe.
- El estudio, de un vistazo
- Quiénes
- Representantes de empresas que analizan la voz con inteligencia artificial
- Cuántos
- once empresas
- Dónde
- Estados Unidos, Canadá, el Reino Unido, Luxemburgo, Alemania, Francia, Australia y Argentina
- Cuándo
- entre febrero y mayo de 2025
- Tipo de estudio
- interviews
- Quién lo hizo
- Instituto Sean Parker para la Voz, Weill Cornell Medical College, Nueva York
- El límite que importa
- Once empresas no dicen qué tan común es esto en el ramo
Dónde se recogieron los datos y a dónde apuntaban los productos
Los datos salieron de países más pobres y los productos se pensaron para países ricos
La voz sirve para detectar enfermedades y también identifica a una persona.
Un representante de una empresa lo dijo sin rodeos: “La desidentificación de la voz es muy problemática… no puedo cambiar la voz porque… entonces no sirve”.
Esa frase encierra el problema. La voz sirve para detectar enfermedades y también identifica a una persona. No se puede separar lo uno de lo otro sin destruir el dato.
Entre febrero y mayo de 2025, investigadores de la Universidad Weill Cornell entrevistaron por videollamada a un representante de cada una de once empresas de salud por voz con sede en Estados Unidos, Canadá, el Reino Unido, Luxemburgo, Alemania, Francia, Australia y Argentina.
El artículo describe un estudio en el que se mencionan reglas que tratan la voz como dato que identifica a la persona: la ley de privacidad médica de Estados Unidos (HIPAA) y el reglamento europeo (GDPR). Aun así, cuatro de las once empresas recogieron datos en países de ingreso medio y bajo, tres de ellos en India. Un representante explicó la diferencia de pagos: “En India son unos pocos dólares al día. En Estados Unidos sería mucho más”. Los productos, en cambio, apuntaban primero a mercados ricos. Una empresa dijo: “Canadá será el primero”.
Dos de las once no habían buscado ninguna revisión ética propia. Una explicó: “Éramos un negocio”. El consentimiento iba de casillas marcadas una por una hasta formularios de una página, exenciones y permisos delegados a quien patrocina el estudio.
El estudio no puede decir qué tan común es todo esto. Son once empresas, contactadas por una red académica, y ellas mismas describieron sus prácticas; nadie las verificó por fuera. Las citas son lo que dijeron, no hechos comprobados. Tampoco se puede concluir que ninguna empresa nombrada o de su país actúe así.
El estudio no informa de ningún fallo judicial ni de ninguna decisión de autoridad en país alguno. Solo señala cómo tratan la voz HIPAA y GDPR. No puede decirle qué permiten sus propias autoridades ni qué puede usted exigir.
Una empresa había cerrado tras no conseguir financiamiento y conservaba grabaciones de voz psiquiátricas. Ningún entrevistado describió un plan para qué hacer con esos datos si la empresa desaparece. La pregunta que queda es esa: cuando una empresa le pida su voz, pregunte quién guardará la grabación, dónde estará y qué pasará con ella si la empresa cierra.
Qué significa para usted
Cuando una empresa le pida su voz, pregunte quién guardará la grabación, dónde estará y qué pasará con ella si la empresa cierra. El estudio no informa de ninguna ley, ningún fallo ni ninguna decisión de autoridad en su país, así que no puede decirle qué permiten sus autoridades ni qué puede usted exigir. Lo que sí puede hacer es pedir esas respuestas por escrito antes de aceptar.
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383
Quién pagó: Los autores declararon haber recibido apoyo financiero; AR recibió una beca Paul B. Beeson Emerging Leaders Career Development Award in Aging (K76 AG079040) del National Institute on Aging y el premio Bridge2AI (OT2 OD032720) del NIH Common Fund. El artículo no indica si los financiadores tuvieron alguna influencia ni menciona equipos o programas cedidos.
No tome esto como consejo médico profesional.
Los hallazgos de otros estudios que aquí se mencionan los conocemos por este documento, que fue el que leímos; no abrimos cada uno de esos estudios.
Versión detalladaLos pasajes copiados del artículo, las ilustraciones y cada fuente con cuánto leímos de ella · 1298 palabras · unos 6 minLeerla →Cerrar
Grabaron su voz en India y África para venderla en países ricos. Dos de once empresas no buscaron revisión ética.
Once empresas que dicen detectar enfermedades con la voz contaron cómo recogen los datos. El estudio no mide cuántas lo hacen así; muestra lo que algunas describieron.
Imagine un teléfono en un consultorio de la India. Una persona habla unos segundos para una prueba de salud. Su voz viaja a un servidor en otro país y sirve para entrenar un producto que se venderá en Canadá o en Europa. Ella cobró unos dólares por ese día.
Ese teléfono aparece en un estudio publicado el 15 de septiembre de 2026 en la revista Frontiers in Digital Health. Los autores son Yuhan, Swain, Zhao, Ramos y Rameau, del Instituto Sean Parker para la Voz del Weill Cornell Medical College, en Nueva York, con el consorcio Bridge2AI-Voice. Entrevistaron a representantes de once empresas que aplican inteligencia artificial a la voz. El trabajo lo pagó el Instituto Nacional del Envejecimiento de Estados Unidos y el fondo común de los Institutos Nacionales de Salud. Una de las autoras asesora a dos empresas del ramo y es fundadora de una tercera, y lo declara en el artículo.
La tecnología se llama biomarcador vocal. Es una característica de la voz, sola o combinada con otras, validada como señal de un desenlace clínico. Las empresas del estudio apuntaban sobre todo a alzhéimer, diabetes tipo 2 y depresión. El artículo no midió si esos productos funcionan ni si mejoran la salud de nadie. Solo preguntó cómo se recogen y se gobiernan los datos. Las entrevistas fueron por Zoom, entre febrero y mayo de 2025, y cada una duró menos de 30 minutos. Participaron once de las 27 empresas contactadas, el 41 por ciento.
El estudio encontró seis patrones. No hubo estándares compartidos para grabar: cada empresa usó sus aparatos, sus formatos y sus etiquetas, y eso vuelve difícil comparar o juntar los datos. La revisión ética varió de un proceso formal completo a nada: dos de las once empresas no buscaron ninguna revisión propia. El consentimiento fue desde casillas detalladas hasta formularios de una página o renuncias justificadas por considerar el estudio de riesgo no significativo. Las voces, según los propios entrevistados, no se pueden anonimizar del todo: la misma señal que sirve de diagnóstico identifica a la persona. Cuatro de las once empresas recogieron datos en países de ingreso bajo o medio, tres de ellas en India, mientras sus productos iban destinados sobre todo a mercados ricos. Y siete de las once ya habían hecho productos comerciales con sus datos de investigación; otras tres estaban en proceso.
Hay límites que importan. Once empresas no dicen qué tan común es esto en el ramo. La muestra se reclutó por la red de un consorcio académico, así que es probable que hayan aceptado las empresas ya acostumbradas a tratar con la ética académica; el resto del sector podría estar peor. Dieciséis empresas no contestaron y no se sabe cómo son. Lo que se reporta son las palabras de los representantes, no hechos verificados por los investigadores. Y varias autoras trabajan en el mismo consorcio que promueve estándares de datos en este campo.
Un dato pone escala al asunto. Una de las empresas dijo haber tenido unos 50,000 participantes en India y 10,000 en cuatro países de África, frente a menos de 1,000 en Europa1. Otra explicó la diferencia de pagos con una frase llana: en India son unos cuantos dólares al día; en Estados Unidos sería mucho más2. Los autores describen este patrón como un ejemplo de la asimetría que Leonelli llama economía epistémica de los datos transnacionales: datos recogidos en países con protección regulatoria más débil que acaban en productos vendidos en mercados ricos3.
La voz es un dato poco común: puede servir como señal clínica y a la vez identifica a la persona, y las dos cosas no se pueden separar sin echar a perder el dato4. Por eso, cuando una empresa quiebra, queda una pregunta sin respuesta. Una de las empresas entrevistadas había cerrado tras no cerrar una ronda de inversión, y conservaba grabaciones psiquiátricas recogidas con consentimientos que daban por hecho que habría alguien cuidándolas5. Ninguna de las once describió un plan para qué hacer con los datos si la empresa desaparece6.
Las reglas que existen tratan la voz como identificador. En Estados Unidos, la Ley de Portabilidad y Responsabilidad de Seguros de Salud la cuenta entre los identificadores que hay que proteger. En la Unión Europea, el Reglamento General de Protección de Datos la trata como dato biométrico de categoría especial cuando se usa para identificar a alguien. Los autores piden a los reguladores que revisen cómo se clasifica la voz, que exijan planes para el destino de las grabaciones y que pidan reparto de beneficios a las comunidades de donde salen los datos78.
Otros trabajos que leímos van en la misma dirección. Una encuesta de la Organización Mundial de la Salud, de la que solo pudimos leer el resumen porque el texto completo está detrás de una suscripción, halló que, entre los 50 Estados miembros que respondieron, 4 tienen una estrategia de inteligencia artificial específica para salud y 7 la están elaborando9; solo 14 han emitido guías sobre las implicaciones éticas de usar esa tecnología en salud10. El Reglamento Europeo de Inteligencia Artificial, según el resumen de otro estudio que tampoco pudimos leer completo, clasifica la mayoría de las herramientas clínicas como de alto riesgo y exige evaluaciones de impacto sobre derechos fundamentales111213. Un tercer resumen, de un sistema de información sobre párkinson desplegado en Alemania, reporta que una revisión experta de conversaciones ya calificadas como buenas encontró errores clínicos críticos que la revisión automática no vio14.
Así lo leemos nosotros. El patrón es este: cuando un servicio se ofrece como comodidad o como pequeño ingreso, lo que la persona entrega a cambio casi nunca se negocia a la vista. Queda dentro del servicio, fuera de su control. Esperamos que muchas de las personas grabadas hayan entendido su participación como una consulta, un favor o unos pesos, sin saber que su voz alimentaría un producto vendido en otro país ni quién lo controla después. Sabríamos que nos equivocamos si esas personas hubieran recibido por escrito para qué se usaría su voz, quién sería su dueño y cómo retirarla.
Y aquí hay algo que usted puede usar. Antes de prestar su voz a una aplicación, a un estudio o a una consulta, pida por escrito tres cosas: para qué se usará, quién será su dueño y cómo se borra. Pida el documento completo, no el resumen, y pregunte qué pasa con su grabación si la empresa cierra. Guarde la respuesta. Es lo que después le permite reclamar. Cuando alguien hable en nombre de la ciencia sobre su voz o la de su comunidad, pregunte quién participó en decidir las reglas y qué poder tuvo esa gente para cambiar algo.
Los autores proponen tres prácticas que al menos una empresa de la muestra ya aplicaba: anotar con qué aparato y en qué lugar se grabó, con qué formato y cómo se asignó la etiqueta clínica; un consentimiento mínimo por escrito, con lenguaje sencillo y opción de retirarse; y revisión ética independiente para toda grabación humana, también para empresas privadas sin universidad detrás. Ninguna de esas prácticas sustituye una atención médica. Son condiciones para que el dato sirva y para que quien lo dio conserve algún control.
Volvamos al teléfono del consultorio. La persona que habló unos segundos quizá no sepa que su voz sigue guardada en algún servidor, ni que nadie describió qué pasará con ella si la empresa cierra. Usted sí puede preguntarlo la próxima vez que le pidan su voz: ¿para qué es, quién la va a cuidar y cómo se borra?
De dónde sale cada dato de contexto, y cuánto leímos de cada documento
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "For one startup the asymmetry was also quantitative: it reported approximately 50,000 research participants in India and 10,000 across four African countries, against fewer than 1,000 in Europe."
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Startups noted that compensation in these settings is lower than it would be in high-income countries (HICs): “In India, it's a few dollars a day. In the U.S., it would be a lot more” (Q41)."
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "This pattern, extracting biological data from countries with weaker regulatory protection and commercializing the resulting products in high-income markets, exemplifies the asymmetry at the heart of what Leonelli terms the epistemic economy of transnational data"
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Voice is an unusual piece of biomedical data: it may serve as a clinically meaningful biomarker while also functioning as a highly identifying piece of information ( 25 )."
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "One interviewed company had halted operations after an unsuccessful fundraising round (“We were in the middle of fundraising and couldn't close the round… so we had to stop”), at which point it held psychiatric voice recordings collected under consent instruments that presupposed a continuing custodian."
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "No interviewee described a plan for data disposition upon dissolution."
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "For regulators and funders, we recommend reassessing the risk classification of voice under HIPAA- and GDPR-style frameworks to reflect its non-anonymizability ( 7 , 8 ) and attaching benefit-sharing and community-engagement expectations to research and products built on data sourced from low- and middle-income settings (theme 5)."
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "We also encourage regulators to require data-disposition plans for biometric datasets, so that the fate of voice recordings is defined before, not after, a company ceases operations."
- Adib K, Letchford N, Dunning HE, Salama N, Tolias Y, De Barros J, et al. (2026). Governance of artificial intelligence for health systems, WHO European Region. Bulletin of the World Health Organization. 10.2471/blt.25.294978 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Of the 50 Member States responding to the survey, 8% (4/50) have a health-specific AI strategy and 14% (7) are developing one."
- Adib K, Letchford N, Dunning HE, Salama N, Tolias Y, De Barros J, et al. (2026). Governance of artificial intelligence for health systems, WHO European Region. Bulletin of the World Health Organization. 10.2471/blt.25.294978 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Only 14 Member States have issued guidelines to address the ethical implications of using AI in health or across sectors."
- Dennstädt F, Hastings J, Windisch P, Jovanovic A, Marić TŽ, Brüningk S, et al. (2026). The EU AI Act: implications and compliance guidance for healthcare facilities. Frontiers in Digital Health. 10.3389/fdgth.2026.1808373 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "The European Union AI Act [Regulation (EU) 2024/1689] establishes the first comprehensive legal framework for artificial intelligence."
- Dennstädt F, Hastings J, Windisch P, Jovanovic A, Marić TŽ, Brüningk S, et al. (2026). The EU AI Act: implications and compliance guidance for healthcare facilities. Frontiers in Digital Health. 10.3389/fdgth.2026.1808373 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "The analysis identifies healthcare as a primary focus of the Act, with most clinical AI tools classified as "high-risk"."
- Dennstädt F, Hastings J, Windisch P, Jovanovic A, Marić TŽ, Brüningk S, et al. (2026). The EU AI Act: implications and compliance guidance for healthcare facilities. Frontiers in Digital Health. 10.3389/fdgth.2026.1808373 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Key results include the definition of mandatory Fundamental Rights Impact Assessments (FRIA), requirements for site-specific technical validation, and the necessity of establishing trust through structured human oversight mechanisms to mitigate automation bias."
- Lange F, Mardi S, Binder T, Reich MM, Odorfer T, Volkmann J. (2026). Real-world use and evaluation of a generative AI chatbot for Parkinson's disease information: a prospective observational study. The Lancet Regional Health - Europe. 10.1016/j.lanepe.2026.101866 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Independent re-review of 100 randomly sampled 'good'-rated conversations identified four (4/100, 4%; 95% CI 1·6-9·8) with clinically critical errors missed by the triage-a conditional false-negative rate within the 'good' stratum, not an overall critical-event rate."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383
Quién pagó: Los autores declararon haber recibido apoyo financiero; AR recibió una beca Paul B. Beeson Emerging Leaders Career Development Award in Aging (K76 AG079040) del National Institute on Aging y el premio Bridge2AI (OT2 OD032720) del NIH Common Fund. El artículo no indica si los financiadores tuvieron alguna influencia ni menciona equipos o programas cedidos.
No tome esto como consejo médico profesional.
Los hallazgos de otros estudios que aquí se mencionan los conocemos por este documento, que fue el que leímos; no abrimos cada uno de esos estudios.
interviews · Frontiers in digital health · the paper, 15 Sep 2026 · free
"I Can't Change the Voice": What 11 Health Startups Said About Your Data
Researchers interviewed 11 companies that build health tools from voice. It cannot say how common the practices they described are.
Short version · the longer version follows, about 6 min
Ask the weeklyAI chat
Ask me about AI in your life: I point you to the stories weeklyAI has published on it, this week and before.
Conversations are saved for as long as weeklyAI exists, to improve the publication. Answers come in the language you write in.
- The study at a glance
- Who
- representatives of vocal biomarker startups
- How many
- eleven startups
- Where
- United States, Canada, United Kingdom, Luxembourg, Germany, France, Australia and Argentina
- When
- February to May 2025
- Kind of study
- interviews
- Who did it
- Weill Cornell Medical College and SUNY Downstate College of Medicine
- The limit that matters
- Eleven companies cannot show how common these practices are across the field.
Approximate numbers of research participants reported by one startup; the Europe figure is given as fewer than 1,000, and these are one company's counts, not the field's.
Where the data came from versus where the products were meant to be sold
Four of the eleven startups collected data in poorer countries, while the products were aimed at rich markets.
One company said a few dollars a day in India, against a lot more in the US.
So when a company asks for your voice, ask who will keep the recording, where it will be stored, and what happens to it if the company closes.
"The de-identification of voice is very problematic," one startup representative told researchers. "I can't change the voice because... then it's not useful."
Behind that sentence sits a tangle. The same recording that might spot Parkinson's disease or depression also carries who you are. The article describes a study in which US health privacy law lists voiceprints among the identifiers it protects. The article describes a study in which Europe's GDPR treats voice recordings as sensitive data when they are used to identify a person. Strip the identity out, and the health signal goes with it.
Researchers at Weill Cornell Medical College interviewed one representative each from 11 vocal biomarker startups. The companies were based in the United States, Canada, the United Kingdom, Luxembourg, Germany, France, Australia and Argentina. The interviews ran from February to May 2025, each under 30 minutes.
Four of the 11 collected voice data in low- and middle-income countries. One reported about 50,000 participants in India and about 10,000 across four African countries. Another said pay in India was "a few dollars a day," against "a lot more" in the US. The products were aimed at rich markets: one company said "Canada will be first."
Two of the 11 had sought no ethics review of their own. One company's table entry noted that it had not sought ethics review and planned to do so through a future university collaboration. Consent ranged from item-by-item checkboxes to waived consent for studies called low risk, and to a one-page form.
Treat the quotes as company claims. The researchers did not check them against records, and the companies described themselves. Only 11 of 27 approached agreed to talk, and they were recruited through an academic network, so the ones who answered may be the ones who think about ethics most.
The study reports no court ruling and no regulator's decision in any country. It cannot tell you what your own authorities allow, or what you may demand of a company that asks for your voice.
Some do release data openly. One put its collection online for anyone to use.
So when a company asks for your voice, ask who will keep the recording, where it will be stored, and what happens to it if the company closes.
What this means for you
What the researchers heard is a set of company claims, not a survey of the field. Eleven startups cannot show how common any practice is, and none of their products was tested for whether it helps or harms a patient. When a company asks for a voice recording, ask what happens to it if the business closes.
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383
Who paid: The authors declared financial support was received; AR was supported by a Paul B. Beeson Emerging Leaders Career Development Award in Aging (K76 AG079040) from the National Institute on Aging and by the Bridge2AI award (OT2 OD032720) from the NIH Common Fund. The article does not say whether funders had any say, and names no equipment or software lent.
Do not take this as professional medical advice.
The findings of other studies mentioned here are known to us through this document, which is the one we read; we did not open each of those studies.
The longer versionThe passages copied from the paper, the pictures, and every source with how much of it we read · 1250 words · about 6 minRead it →Close
Startups Are Recording Voices Abroad to Sell Health Tools at Home, and Often No One Reviews the Ethics
A qualitative study of eleven voice-biomarker companies found uneven ethics review, thin consent, and data collected in poorer countries for products aimed at rich ones. It cannot say how common any of this is.
Consider a smartphone app that asks you to count backward for thirty seconds, or to hold a vowel. It records your voice. Artificial intelligence can then be used to screen for, diagnose, and monitor conditions including Parkinson's disease, Alzheimer's disease, depression, and heart failure. That is vocal biomarker technology: a person's voice, measured by artificial intelligence, used as a sign of health or illness1.
A study published on 15 September 2026 in *Frontiers in Digital Health* interviewed representatives from eleven such startups. The authors are affiliated with Weill Cornell Medical College and SUNY Downstate College of Medicine, and one author was supported by awards from the National Institute on Aging and the NIH Common Fund. The interviews took place by video call between February and May 2025, each lasting under thirty minutes. Three of the companies were headquartered in the United States, two in Canada, and one each in the United Kingdom, Luxembourg, Germany, France, Australia, and Argentina. Twenty-seven companies were approached; eleven agreed to take part, or 41 percent.
Six themes emerged. First, the companies used no shared standard for recording audio, so a phone in one pocket and a laptop on a desk produced different data. Second, ethics review varied widely. Two of the eleven startups had sought no ethics review of their own. Among the nine that did, three used commercial review boards, which the authors suggest may deserve scrutiny because such boards offer expedited approval23. Third, consent ranged from item-by-item checkboxes to a single page, and one company said its studies sometimes received a waiver of consent on the grounds that they posed "non-significant risk," a judgment the authors question given what voice reveals about identity4. Fourth, voice cannot truly be anonymized: the same qualities that make it medically useful also make it recognizable5. Fifth, four of the eleven companies collected data in low- and middle-income countries, three of them in India, while their products were meant mainly for wealthy markets67. Sixth, seven of the eleven had built commercial products from their research data, and three more were in the process.
The study is a description, not a measurement. Eleven companies cannot tell you how common these practices are across the field, and the authors say so plainly8. Recruitment ran through an academic network, which likely attracted companies already thinking about ethics. The practices described came from the companies themselves and were not independently checked. The researchers also belong to a consortium that promotes data standards in this field, which they acknowledge may have shaped what they heard and how they read it.
The gap between what a company records and what a person understands is not unique to voice. Think of the terms-of-service screen on a new phone: a wall of text, a single button, and no memory afterward of what was agreed. The study found consent practices that ranged from granular opt-in to a deliberately shortened one-page form, and one company described a compromise between brevity and its data protection officer's wish for length. One startup said its studies sometimes received a waiver of consent altogether4.
What makes voice different is that it cannot be separated from the person. One representative put it directly: de-identifying a voice is very problematic, because changing the voice makes it useless5. Another noted that a breach of a voice database would be especially severe for exactly that reason9. This is the trade-off at the center of the field: the signal that carries diagnostic value is the same signal that carries identity, and you cannot keep one without the other.
The geography is where the study becomes harder to look away from. Four of eleven startups collected data in poorer countries; one reported roughly 50,000 participants in India and 10,000 across four African countries, against fewer than 1,000 in Europe10. Compensation in those settings is lower: one company said a few dollars a day in India, against much more in the United States11. The products, meanwhile, were aimed primarily at high-income markets7. The authors describe this as extracting data where regulation is weaker and selling the results where the money is12.
What happens when a company fails? One interviewed startup had halted operations after an unsuccessful fundraising round, and it held psychiatric voice recordings collected under consent forms that assumed a continuing custodian. No interviewee described a plan for what happens to the data if the company dissolves1314. For a person who donated a voice sample, that is the question with no answer in the study.
The authors' recommendations are concrete: reassess how voice is classified under privacy frameworks like HIPAA and GDPR, attach benefit-sharing expectations to data collected in poorer countries, and require companies to plan for what happens to recordings when they close1516. They also call for independent ethics review for every company collecting human voice data, including those with no university affiliation.
Here is how we read it. The pattern is familiar from other corners of the digital economy: the party with money and organization sets the terms, the party with less power clicks agree, and the know-how and the recordings end up on the side that paid. We would expect voice-recording arrangements in poorer regions to be structured by the buyer, with local partners handling paperwork and payment set at local rates. We would be wrong if those deals showed community groups co-designing terms, sharing in later profits, and holding copies of the recordings themselves. If a clinic, university, or employer invites you to donate your voice, ask in writing who owns the recording, where it is stored, and what your community gets if a product is sold.
That is a disagreement about the technology, not about what these companies did.
And we noticed a boundary: the study tells us nothing about what courts or regulators in your country have decided about voice data. It notes only that HIPAA and GDPR treat voice as identifying. If you want to know what your own authorities require, you will have to ask them.
Other work in this area reaches compatible conclusions. A survey of governance across the WHO European Region found that only 8 percent of responding member states had a health-specific AI strategy, and 14 had issued guidelines on the ethics of AI in health; we could read only the summary of that study1718. A separate analysis of the European Union's AI Act found that most clinical AI tools are classified as high-risk and that healthcare facilities lack practical frameworks to move from legal theory to compliance; we could read only the summary1920.
For the reader, the practical question is not whether this particular study proves anything about your country's startups. It does not. The question is what you can ask before you speak into an app. Ask who reviews the ethics of the recording. Ask what happens to your voice if the company closes. Ask whether the consent form lets you choose item by item. Those questions cost nothing, and the study suggests that in at least some companies, no one has been asked them yet.
The phone in your hand can record a voice in seconds. The paper trail that should follow it takes longer, and someone has to ask for it.
Where each piece of context comes from, and how much of it we read
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "Voice is an unusual piece of biomedical data: it may serve as a clinically meaningful biomarker while also functioning as a highly identifying piece of information ( 25 )."
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "At the other end, two of the eleven startups had sought no ethics review of their own"
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "Among the nine that did obtain review, three used commercial or central IRBs, which may itself warrant scrutiny given that such entities offer expedited approval."
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "At one end, one organization reported that its studies received waivers of consent on the grounds of being “non-significant risk,” a categorization that warrants scrutiny given the biometric properties of voice data and the risk of re-identification."
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "Representatives described voice as inherently identifiable and pointed to the difficulty of stripping identity from it: “The de-identification of voice is very problematic… I can't change the voice because… then it's not useful” (Q30)."
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "Four of the eleven startups collected data from low- and middle-income countries (LMICs), three of them in India."
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "Yet while these companies sourced data in LMICs, their products were intended primarily for HIC markets (Q39–Q40)."
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "First, our sample of eleven startups characterizes patterns present among these companies; it cannot establish the prevalence of these practices among startups across the vocal biomarker field, and our findings may not be generalizable."
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "Another organization echoed this concern, noting that a breach of a voice database would be especially severe precisely because the data are intrinsically tied to individual identity."
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "For one startup the asymmetry was also quantitative: it reported approximately 50,000 research participants in India and 10,000 across four African countries, against fewer than 1,000 in Europe."
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "Startups noted that compensation in these settings is lower than it would be in high-income countries (HICs): “In India, it's a few dollars a day. In the U.S., it would be a lot more” (Q41)."
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "This pattern, extracting biological data from countries with weaker regulatory protection and commercializing the resulting products in high-income markets, exemplifies the asymmetry at the heart of what Leonelli terms the epistemic economy of transnational data"
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "One interviewed company had halted operations after an unsuccessful fundraising round (“We were in the middle of fundraising and couldn't close the round… so we had to stop”), at which point it held psychiatric voice recordings collected under consent instruments that presupposed a continuing custodian."
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "No interviewee described a plan for data disposition upon dissolution."
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "For regulators and funders, we recommend reassessing the risk classification of voice under HIPAA- and GDPR-style frameworks to reflect its non-anonymizability ( 7 , 8 ) and attaching benefit-sharing and community-engagement expectations to research and products built on data sourced from low- and middle-income settings (theme 5)."
- Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "We also encourage regulators to require data-disposition plans for biometric datasets, so that the fate of voice recordings is defined before, not after, a company ceases operations."
- Adib K, Letchford N, Dunning HE, Salama N, Tolias Y, De Barros J, et al. (2026). Governance of artificial intelligence for health systems, WHO European Region. Bulletin of the World Health Organization. 10.2471/blt.25.294978 — only the abstract - the full text could not be fetched — the passage: "Of the 50 Member States responding to the survey, 8% (4/50) have a health-specific AI strategy and 14% (7) are developing one."
- Adib K, Letchford N, Dunning HE, Salama N, Tolias Y, De Barros J, et al. (2026). Governance of artificial intelligence for health systems, WHO European Region. Bulletin of the World Health Organization. 10.2471/blt.25.294978 — only the abstract - the full text could not be fetched — the passage: "Only 14 Member States have issued guidelines to address the ethical implications of using AI in health or across sectors."
- Dennstädt F, Hastings J, Windisch P, Jovanovic A, Marić TŽ, Brüningk S, et al. (2026). The EU AI Act: implications and compliance guidance for healthcare facilities. Frontiers in Digital Health. 10.3389/fdgth.2026.1808373 — only the abstract - the full text could not be fetched — the passage: "There is currently a lack of practical operational frameworks to help healthcare facilities transition from legal theory to clinical compliance."
- Dennstädt F, Hastings J, Windisch P, Jovanovic A, Marić TŽ, Brüningk S, et al. (2026). The EU AI Act: implications and compliance guidance for healthcare facilities. Frontiers in Digital Health. 10.3389/fdgth.2026.1808373 — only the abstract - the full text could not be fetched — the passage: "The analysis identifies healthcare as a primary focus of the Act, with most clinical AI tools classified as "high-risk"."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383
Who paid: The authors declared financial support was received; AR was supported by a Paul B. Beeson Emerging Leaders Career Development Award in Aging (K76 AG079040) from the National Institute on Aging and by the Bridge2AI award (OT2 OD032720) from the NIH Common Fund. The article does not say whether funders had any say, and names no equipment or software lent.
Do not take this as professional medical advice.