CAMBRIDGE, United Kingdom, Oct. 14, 2026 (GLOBE NEWSWIRE) -- Today, Speechmatics releases Oak, a medical speech recognition model built on new architecture. Oak transcribes clinical conversations as speakers switch languages mid-sentence, keeping drug names, doses, and diagnoses highly accurate.
More than one in five people in the US speak a language other than English at home*, and clinicians mirror that trend with one study showing that 71.6% of first-year US resident physicians speak a second language**. Clinical conversations switch between speakers and languages constantly, often mid-sentence. Yet existing speech recognition models do not accurately reflect that clinical reality. Most providers build medical models for English only, and general models often get medical terms wrong. Oak was developed to capture real multilingual medical conversations.
Model evaluation also needs to change to prove real value. Most speech recognition providers report accuracy as word error rate (WER), which gives the same ‘value’ to every word. Mistranscribing "the" and "amlodipine" does not present the same risk. One is a typo, and the other puts the wrong drug on a patient's record. In a setting where errors have a huge impact, WER is not rigorous enough. Instead, Speechmatics measures Oak on medical term recall specifically, which focuses on the share of correctly transcribed drug names, conditions, procedures and side effects.
In an internal Speechmatics study, the Oak model accurately transcribed 91% of all medical terms in English and 10 additional commonly used languages, tested on both clean audio and recordings with realistic background noise. The eight other providers missed or mistranscribed an average of 25.7% of medical terms - roughly one drug name, condition, or side effect in every four.
Spanish is the most widely spoken language after English in US homes, with 42 million speakers***. In Spanish, Oak transcribed 96.3% of medical terms, against 86.8% on average for the other eight providers. For every 100 drug names, conditions, and side effects spoken in a consultation in Spanish, Oak misses about four, while the others miss 13 on average, more than three times as many.
Oak transcribes a conversation across languages and speakers, switching mid-sentence with no language set in advance. It supports 55 languages and can transcribe an hour of audio accurately within 18 seconds, allowing for more optimized workflows.
Oak was developed by Speechmatics and can be deployed in the cloud or self-hosted within a healthcare organization's own environment. For self-hosted deployments, Speechmatics uses NVIDIA Dynamo-Triton to serve Oak, giving organizations greater control over where sensitive patient audio is processed. Speechmatics also uses NVIDIA Nsight Systems for performance profiling and optimization, alongside quantized NVIDIA CUDA kernels to improve inference efficiency.
Oak builds on years of experience deploying speech recognition technology in high-stakes medical environments. Humetrix switched to Speechmatics for multilingual patient intake and ran it for patients from 97 countries at the Paris 2024 Olympics. Sully.ai uses Speechmatics for clinical scribes and AI receptionists serving more than 500 provider customers and in the UK, tens of millions of NHS 111 calls per year are transcribed using Speechmatics via partner Content Guru.
Katy Wigdahl, CEO of Speechmatics, said, "Our mission is to understand every voice and the risk of getting that wrong in a medical environment is huge. A consultation can move from Spanish to English and back in one sentence, with a drug name in the middle. That switch is the precise moment a clinician can least afford an error.”
"Getting clinical terms right in every language is one of the hardest problems in speech recognition. Oak shows it can be solved when the model is built for how clinicians and patients actually speak, and we’re proud to see it set a new bar for accuracy where it matters most.”
Notes to Editors:
- Speechmatics is HIPAA-compliant, SOC 2 Type II certified, ISO/IEC 27001:2022 accredited, and GDPR-compliant.
- Speechmatics supports multiple endpoints where data can be processed to comply with sovereign requirements.
- Speechmatics runs across cloud, private cloud, on-premises and edge environments, so healthcare providers can keep patient audio inside their own infrastructure, and its data processing environments around the world help customers meet compliance, regulatory and sovereignty obligations.
Benchmarking methodology:
Speechmatics internal medical speech recognition evaluation, 29 September 2026. Oak tested against AWS, AssemblyAI, Deepgram, ElevenLabs, Google, Microsoft, OpenAI and Soniox across 19 medical test sets in 11 languages (one language per test set), using the same audio and scoring for every provider, covering both clean audio and audio with realistic background noise. All providers ran on default settings with no custom vocabulary, using each provider's best available model per language, including dedicated medical models where offered.
About Speechmatics
Speechmatics builds the most accurate and inclusive speech recognition in the world, trusted by enterprises and developers across financial services, media, contact centers and government for unmatched language coverage, accuracy, and flexibility.
Data sources:
- US Census Bureau, 2017-2021 ACS language use tables (22% speak a language other than English at home; 500+ languages)*
- US Census Bureau, Language Use in the United States: 2019 (ACS-50) (42 million people spoke Spanish at home, 2019)***
- Tijerina M, Utama R, Harris M, Ortega P. Specialty trends in first-year resident physician multilingualism from 2013 to 2024. Academic Medicine, August 2026: https://pubmed.ncbi.nlm.nih.gov/41739692/**
A photo accompanying this announcement is available at https://www.globenewswire.com/NewsRoom/AttachmentNg/f96a4f23-0138-4753-890c-ad31d4956f54
Contact:
Mieke Smith
Mieke.smith@speechmatics.com
01223 794497
