Amazon’s machine learning transcription service aims to ease docs’ tasks

Amazon Web Services is rolling out an electronic health record-supported machine learning transcription service that uses speech recognition applications to ease physician documentation.

The product is Amazon Transcribe Medical, which automatically translates audio streams into medical speech, enabling affordable, secure and accurate note taking for clinical staff, researchers and other stakeholders.


Cerner, for example is using the product in an initial development of a digital voice scribe that automatically listens to clinician and patient interactions and captures the conversation in text form. The service enables developers to add medical speech-to-text capability to their applications.

Amazon is positioning Transcribe Medical as a tool to ease physician and researcher burnout. The company notes that clinicians can spend six hours a day on top of their medical tasks writing notes for entry into the electronic heath record, resulting in additional stress, a poorer level of care and rushed patient visits.

To avoid transcription hassles, Transcribe Medical offers an app programming interface that integrates with any voice-enabled application and any device that has a microphone. The company further offers a fully managed service that requires no provisioning or management of servers. A user needs only call the public application programming interface and start passing an audio stream to the service via a secure WebSkocket connection. Experience in machine learning is not required.

George Seegan, senior data scientist at Amgen, a biotechnology company studying biological mechanisms to find new therapies, says Transcribe Medical helps him quickly find the information he needs.

“In pharmacovigilance, we want to accurately review recorded calls from patients or healthcare providers to identify any reported potential side effects associated with medicinal products,” Seegan explains. “Amazon Transcribe Medical produces text transcripts that allow us to extract meaningful insights about medicines and any reported side effects. In this way, we can quickly detect, collect, assess, report and monitor adverse effects to the benefit of patients.”

Many physicians have employed medical scribes to handle medical note taking, but that is an expensive option, and some patients may not feel comfortable talking candidly about their issues with a scribe in the room.

Further, some physicians don’t view documentation software as being better than having a scribe.

Also See: Amazon launches NLP service to process unstructured text

That’s because existing front-end dictation software requires clinicians to take training, which is time-consuming and possibly expensive, or having to speak unnaturally, such as calling out punctuation, which is disruptive and inefficient.

To reduce note taking, providers may send physicians’ voice recordings to manual transcription services, but the turnaround time can be slow—sometimes, as long as three days.

Physicians face another hurdle when they move into the life sciences arena in preparation for practicing precision medicine. Pharmaceutical companies increasingly need to collect real-world evidence of their medications’ efficacy, or whether medications could cause possible side effects. However, real-world evidence is often acquired during phone calls and the calls then need to be transcribed.