Standard phone speech recordings that neurology drug trials can use as a measure
TL;DR Vowelwatch supplies drug-trial sponsors with a validated set of phone speech tasks in Indian languages. Trial participants with Parkinson's, ALS or similar conditions record at home between site visits, and the sponsor gets clean, comparable speech measures. Sponsors and CROs pay per study. It sells measurement, not diagnosis.
Neurology trials struggle to measure slow change. Clinic rating scales are coarse, they depend on who's doing the rating and they only happen at site visits. Speech is one of the few functions a phone can record at home, and it changes in Parkinson's, ALS and other conditions.
India is an obvious place to run such trials. An estimated 771,000 people had Parkinson's in 2019, with crude prevalence up 105.9% since 1990[1]. Not everything is hunky-dory though. Speech tools built in English don't transfer to Hindi, Tamil or Bengali speakers, so sponsors running Indian sites either skip speech measures or collect data they can't pool.
What it does, in regulatory terms - a data-collection tool for approved clinical trials, measuring speech in already-diagnosed participants. It does not diagnose or guide treatment. The trial runs under the New Drugs and Clinical Trials Rules 2019 with ethics approval[2]. Selling it for routine clinical use later would need a CDSCO device licence[3].
Participants get an app with fixed tasks in their own language - hold a vowel, count, read a passage and describe a picture. The app checks background noise before each recording. Vowelwatch extracts measures such as loudness, pitch range and speaking rate, and delivers them to the sponsor with a record of how each was validated.
The seed proposed diagnosing disease from the voice. The largest test I found, using telephone recordings from 6,531 people, reached only about 67% sensitivity and specificity for Parkinson's[4]. That is far too weak to diagnose anyone. Measuring change within a diagnosed person across a trial is a job you can actually do.
Trial participants record and the sponsor pays, usually through a contract research organisation. Pricing in this field is set per study or per participant and nobody publishes it. Academic neurology groups running investigator-led studies are a smaller second buyer, often grant-funded. Expect the first deal to be a paid add-on to one study rather than a platform licence.
The US FDA finalised guidance on digital health technologies for remote data collection in trials in December 2023[5], which gives sponsors a framework for tools like this. Small studies are testing phone speech tasks in ALS, with weekly home assessments over 24 weeks[6]. Speech models now handle Indian languages on cheap phones.
Nobody publishes a figure for spending on speech measures in Indian trials. The narrowest real number is the 771,000 people with Parkinson's in India[1], a pool that sponsors recruit from.
Foreign voice-biomarker companies list clinical trials among their target customers[7]. That's an adjacent signal, and there's no Indian revenue published behind it.
One US firm offers a vocal-biomarker platform for providers and clinical trials[7]. Another sells voice APIs for mental, cognitive and respiratory fitness, mainly to employers and developers[8]. Neither publishes validated tasks in Indian languages. I found no Indian firm doing this, though my search was limited. The gap is local-language validation that sponsors can trust.
| Difficulty | Hard (55/100) |
| Starting capital | ₹1.5–3 Cr |
| Time to launch | 15–24 months |
The hard part is evidence. You must show each measure repeats reliably and moves with disease, in each language, before a sponsor will touch it. Sales cycles with sponsors are long and slow, and the first deal usually needs published data from more than one site.
Yes. Trials are tightly regulated, and clinical use later would bring device rules too.
Building this outside India? The categories above usually have a local equivalent, so check with your own regulator.
AI does the measuring here, so the uses are about trust in the numbers.
Don't use these measures to diagnose anyone or to guide treatment. They track change in already-diagnosed participants, and only within the languages you've validated.
Want to work out the AI layer for Vowelwatch? Talk to Yash →
Here's what can go south -
This is only for a team with speech-science depth, a movement-disorder neurologist and links to CROs. If you want to sell voice-based diagnosis to the public then I'd steer well clear.
Prove one thing first - in a 50-patient study, the measures repeat week to week and track a clinic rating over six months.
Plan for roughly ₹1.5–3 Cr, mostly on validation studies, speech-science talent and long sponsor sales cycles.
Each trial needs CDSCO and ethics approval through the sponsor. Selling the measures for routine clinical use later would need a CDSCO device licence.
Yes, most sponsors run global trials. Follow the US FDA digital health technology guidance and your local trial rules from the start.
The free Vowelwatch validation kit has the line-by-line capital breakdown, the effort scoring, the trial-rules and data-protection checklist, a 14-day validation sprint with pass/fail targets, a 90-day plan, 10 interview questions for sponsors, CROs and neurologists, and 5 assumptions to test. Download the kit →
Thinking about building Vowelwatch or something close to it? The make-or-break work is publishing validation data good enough for a sponsor to bet a trial on, and I can help you plan the study and the money behind it. If you need a hand, you know where to find me. Talk to Yash →
For illustration only. Names are fictional and ideas hypothetical; sourced facts are real but may be out of date. Nothing here is legal, financial or professional advice, and Yash Khatri is not personally liable for any use of it. Read the full disclaimer.