Synthio Labs has published DOSE, the first public benchmark of how accurately text-to-speech systems pronounce drug names, testing nine commercial models on 274 medicines. General-purpose models passed just 63.1% to 80.3% of names and fell sharply on newly approved drugs — a gap the company frames as a patient-safety issue.
On September 17, 2026, San Francisco-based Synthio Labs, a clinical-grade voice and agentic AI company for pharma, published DOSE (Drug-name Oral Synthesis Evaluation) — the first public benchmark of how accurately text-to-speech systems pronounce drug names. DOSE tests nine commercial models on 274 drug names, 146 of them newly approved, each read inside a real clinical sentence. The headline finding: sounding natural and saying the medicine correctly are different skills.
General-purpose models passed between 63.1% and 80.3% of drug names overall — up to one in three mispronounced — and every model fell on recently approved medicines. ElevenLabs' eleven_v3 dropped from 93.0% on established names to 67.1% on new ones, and Google's Gemini TTS fell from 89.1% to 61.6%. Microsoft Azure passed fewer than half of generic (INN) names, and one system spelled 'Xofluza' letter by letter instead of pronouncing it. Each pronunciation is scored from 0 to 5 against verified reference pronunciations, with a score of 4 or higher counting as a pass. Synthio's own RxPronounce model led the field, passing 91.2% of names overall and 87.0% of newly approved names — 10.9 points ahead of the next-best system.
'Voice models are sold on how human they sound. A model can sound human and still mispronounce the drug name, and in pharma that is the failure that matters,' said Rajashekar Vasantha, co-founder and CTO of Synthio Labs. He noted that sound-alike drug name confusion is a patient-safety category already tracked by the WHO, ISMP, and FDA — and that voice AI adds a new speaker to that problem that nobody had measured. Co-founder and CEO Supreet Deshpande tied the failures to training data: 'Every model handles metformin, a legacy diabetes drug, but they struggle with the molecule approved last quarter, exactly the name a launch team or a patient most needs spoken correctly.'
The full dataset, including reference pronunciations, scores, and audio, is published on Hugging Face. Synthio says its platform is already deployed live at several global top-10 pharma companies and is available through AWS Marketplace. The benchmark lands as voice AI spreads through healthcare workflows, positioning pronunciation accuracy as a measurable, procurement-grade requirement for clinical voice AI rather than a polish detail.
