Tobi Olatunji, a medical doctor in Nigeria, knows the challenges of working in Africa’s bustling hospitals. Having seen over 30 patients daily while employed at one of West Africa’s largest facilities, he knew that something had to give. That’s where his background as a machine-learning scientist came in handy.
To provide a solution, Olatunji founded Intron Health, which offers software that significantly reduces the time spent on paperwork. This digital record-keeping also speeds up medical research – a feat that is difficult to achieve using paper records.
While pursuing his medical practice, Olatunji earned advanced degrees in medical informatics from the University of San Francisco and in computer science from Georgia Tech. He started working as a machine-learning scientist in the U.S., coding at night and on weekends to help digitize Africa’s hospitals.
However, the process wasn’t entirely smooth-sailing. During the pandemic, the first few doctors who tested Intron Health’s software still found it challenging to use. The natural pairing of medical terminology and thick African accents produced dreadful results with most speech-to-text programs. Thus, Olatunji made the tough choice of investing in natural language processing (NLP) and speech recognition technology.
The Intron team evaluated several commercial and open-source speech recognition frameworks before deciding to build with NVIDIA NeMo, a software framework for text-based generative AI. Additionally, the resulting models were trained on NVIDIA GPUs, cutting the time health workers spent on paperwork by as much as six times across multiple African countries where Intron conducted an ongoing study.
Furthermore, the app has more than 92% accuracy across more than 200 African accents. It even captures doctors’ dictated messages, freeing them from extensive documentation work. Intron refreshes its models every other month since they need high-quality audio data, which is often challenging to obtain.
Intron’s impressive efforts don’t end there. The company developed an app to capture sound bites of medical terms spoken in different accents, gathering over a million clips from more than 7,000 people across 24 countries, including 13 African nations. This rich dataset is essential in training high-quality NLP and speech recognition models, with parts of it released as open source to support African speech research.
Furthermore, Intron launched AfriSpeech-200, a developer challenge that encourages research using the data it gathered. Much of this stems from the fact that very little research exists on speech recognition for African accents in a clinical setting. Additionally, the Bio-RAMP Lab aims to foster diversity and inclusion in medtech. Comprising minority researchers working at the intersection of AI and healthcare, the group already has half a dozen papers under review at major conferences.
Finally, Intron goes beyond providing software solutions; it also helps hospitals in Africa find creative ways to acquire the hardware they need. By digitizing healthcare data, we unlock a whole new world for research into areas like predictive models that can serve as early warning systems for epidemics – a critical need in today’s world.