Along with the launch of electronic prescriptions, which is part of the government's medical digital transformation initiative, and the mandatory use of My Number cards as insurance cards, progress is being made in establishing an electronic infrastructure for medical information.
Based on this, Mitsubishi Electric Digital Innovation Corporation offers "AnyCOMPASS," a next-generation communication service for dispensing pharmacies that aggregates various patient medical data and allows it to be shared and utilized by pharmacies and the local community. The "AI Assistant" function on this service uses AmiVoice's voice recognition to link with generative AI to support medication instructions and medication history creation. We asked about the background and effects of adopting the AmiVoice API in a field where medical terminology recognition and real-time performance are required.

Issues/background

Developing services for the medical industry
Real-time performance and security are issues

The "AI Assistant" in "AnyCOMPASS" is a function that converts conversations between patients and pharmacists into text using voice recognition and utilizes generative AI to support medication instructions and the creation of medication histories. Therefore, it is necessary to accurately capture the conversations between patients and pharmacists in real time and reflect them in medication histories.
Furthermore, as the service handles medical information, security must be guaranteed, and conditions include the disclosure of limited URLs and encrypted communication using secure protocols.However, the demo program that was being developed at the time of the 2024 exhibition was implemented using a free voice recognition API, which posed security issues.

We were struggling to find a service that could provide both security and real-time performance, but then we found AmiVoice on Melphin, our on-premise pharmacy package.As we use a medical package product, we learned about AmiVoiceAPI, a voice recognition API for development. We decided to adopt AmiVoice API because it can provide the real-time performance and security that we require, has a medical engine with an extremely high recognition rate for medical terminology in Japanese, and allows for easy integration with our services via API.

Introduction Results

Excellent recognition of medical terminology, highly rated by pharmacists

AmiVoice's Japanese speech recognition has been researched in-house for many years, resulting in extremely high accuracy in converting text. The "medical engine"'s speech recognition accuracy for medical terminology is particularly outstanding, allowing conversations with patients, where mishearing or omissions are unacceptable, to be accurately converted into text and displayed on the screen in real time. The high accuracy of the speech recognition results also increases the accuracy of the automatic medication history generation by the linked generation AI, earning it extremely high praise from pharmacists who use the service.
Another benefit is that the AmiVoiceAPI management screen allows us to see the usage time and charges for each pharmacy, which means we can quickly calculate the charges to the pharmacy.

Future prospects

Utilizing generative AI to improve pharmacist work
Developing services that can provide further support

AnyCOMPASS plans to continue strengthening its services that utilize generative AI.
In fiscal 2025, we will follow the remarkable development of generative AI, including AI agents and multimodal support, and develop services that can further support the work of pharmacists. As voice will also be an important element as part of multimodal support, we look forward to the continued evolution of the AmiVoice API.

*AnyCOMPASS® and the AnyCOMPASS logo are registered trademarks of Mitsubishi Electric Digital Innovation Corporation.
*Melphin is a registered trademark of Mitsubishi Electric Digital Innovation Corporation.