While speech recognition is becoming integrated into nearly every aspect of modern life, significant gaps remain: people who speak minority languages or have speech disorders such as strong accents or stuttering typically have difficulty using speech recognition tools that can control applications, transcribe or automate tasks.
Tobi Olatunji, founder and CEO of clinical speech recognition startup Intron Health, wants to fill this gap. He claims that Intron is the largest clinical database in Africa, and its algorithms are trained on 3.5 million audio clips (16,000 hours) from more than 18,000 contributors (mostly medical professionals) representing 288 accents across 29 countries. Having most of the contributors from the medical field ensures that medical terms are pronounced correctly and recognized accurately for the target market, Olatunji says.
“We've already trained on many African accents, so the baseline performance for their access is likely to be much better than any other service they use,” he said, adding that with growing data from Ghana, Uganda and South Africa, the startup is confident about deploying the model in those regions.
Olatunji’s interest in healthtech stems from two components of his experience: First, he trained and practiced as a doctor in Nigeria, where he witnessed the inefficiencies in the system in that market – the amount of paperwork that had to be filled out and how difficult it was to keep track of it all.
“When I was a doctor in Nigeria years ago, as a medical student and even now, I easily get frustrated by repetitive tasks that are not worth the human effort,” he says. “A simple example is having to write the patient's name for every test request. A simple example is, let's say I'm seeing a patient and they need to get a prescription and get a test. I have to write every request for the patient by hand. It's just frustrating for me to have to repeat the patient's name, age, date, etc. over and over again on every form… I'm always thinking how can we do it better, how can we make the doctor's job easier, how can we offload some tasks to another system so that they can spend their time on things that are very valuable.”
These questions drove him to take the next step in his life: he moved to the US and earned a master's in health informatics from the University of San Francisco, then a master's in computer science from the Georgia Institute of Technology.
He then gained experience at several technology companies, including as a clinical natural language programming (NLP) scientist and researcher at Enlitic, a San Francisco Bay Area company where he built models to automatically extract information from radiology text reports, and as a machine learning scientist at Amazon Web Services. At both Enlitic and Amazon, he focused on natural language processing in healthcare, building systems to improve hospital operations.
Through these experiences, he began to develop ideas about how things developed and used in the United States could be used to improve healthcare in Nigeria and similar emerging markets.
Founded in 2020, Intron Health's initial aim was to digitize hospital operations in Africa through an electronic medical record (EMR) system, but implementation was difficult: It turned out that doctors preferred writing by hand over typing, Olatunji said.
From there, he began looking for ways to improve a more fundamental problem: how to make doctors' basic data entry, or writing, more efficient. The company first looked at third-party solutions that would automate tasks like note-taking and incorporate existing speech-to-text technology into his EMR program.
However, frequent transcription errors created many problems. Olatunji realized that strong African accents and complex medical terminology and name pronunciations made the implementation of existing foreign transcription tools impractical.
This led to the creation of Intron Health's speech recognition technology, which can recognise African accents and also integrate into existing EMRs. To date, the tool has been deployed in 30 hospitals across five markets, including Kenya and Nigeria.
Some positive results have been immediate: In one case, Intron Health helped one of West Africa's largest hospitals reduce the wait time for radiology results from 48 hours to 20 minutes, Olatunji says. Such efficiency is critical in healthcare delivery, especially in Africa, where doctor-to-patient ratios remain among the lowest in the world.
“Hospitals are already spending a lot of money on equipment and technology. It's important to ensure that these technologies are applied. We can provide value to drive adoption of EMR systems,” he said.
Going forward, the startup is exploring new areas of growth with the backing of a $1.6 million pre-seed round led by Microtraction with participation from Plug and Play Ventures, Jaza Rift Ventures, Octopus Ventures, Africa Health Ventures, OpenseedVC, Pi Campus, Alumni Angel, Baker Bridge Capital, and several angel investors.
On the technology side, Intron Health is perfecting noise cancellation, working to ensure the platform works well in low bandwidth environments, enabling transcription of multi-speaker conversations, and integrating text-to-speech capabilities.
Olatunji said the plan is to add intelligence systems and decision support tools to operations such as prescriptions and lab tests. These tools will help doctors reduce errors and ensure proper care for patients as well as speed up their work, he added.
Intron Health is one of a growing number of generative AI startups in healthcare, including Microsoft's DAX Express, which reduces administrative work for clinicians by generating notes within seconds. The emergence and adoption of these technologies comes as the global speech recognition market is projected to reach $84.97 billion by 2032, growing at a CAGR of 23.7% from 2024, according to Fortune Business Insights.
Beyond building speech technology, Intron also plays a key role in speech research in Africa, recently partnering with Google Research, the Bill & Melinda Gates Foundation, and PATH's Digital Square to evaluate popular large-scale language models (LLMs) such as OpenAI's GPT-4o, Google's Gemini, and Anthropic's Claude in 15 countries to identify the strengths, weaknesses, and risks of bias and harm in LLMs — all in an effort to ensure culturally tailored models are available in clinics and hospitals across Africa.