Shift Plus Inc., together with local governments, plans, develops, and delivers eRex, a speech recognition and transcription service that government and public-sector employees can use naturally in their day-to-day work — even in environments operating under LGWAN (the Local Government Wide Area Network). To explore how eRex addresses the level of accuracy and real-world usability required in municipal operations, as well as the role played by the AmiVoice API, we interviewed Ms. Manami Nagata of the Municipal Platform Department about the background behind its adoption and the results achieved after implementation.

Issues/background

Increased burden of creating meeting minutes, etc., and limitations on tools in local government environments.

In local government operations, audio data is generated in various daily situations such as council meetings, committee meetings, conferences, discussions, and customer service interactions. However, the time required to create meeting minutes and notes has been a long-standing problem. Furthermore, even with recordings, the need for repeated listening does not lead to increased efficiency, and outsourcing the process results in high costs and long lead times.

On the other hand, the reality was that there were very few speech recognition services (tools) available within the LGWAN environment that also possessed practical-level Japanese accuracy. The abundance of proper nouns and system names specific to local governments was also a factor that existing services could not adequately handle.
To address these issues, we decided to collaborate with local governments on the planning and development of eRex, aiming for a system that "local government employees can operate themselves and use naturally within their normal workflow."

 

Key Drivers of the Adoption

High recognition accuracy, especially in the recognition of administrative terminology.

At eRex, after comparing several domestic and international engines, the deciding factors for choosing AmiVoice were: "the high accuracy of Japanese recognition in the AmiVoice API," "its strength in handling government-specific expressions and proper nouns," "the abundance of business-oriented features such as the word registration API and speaker dialization," and "the ease of implementation and operation as an API."

In particular, local government work frequently involves non-common terms such as "personal names," "place names," "system names," and "project names."
The ability to pre-register terms and proper nouns for each municipality using the "Word Registration API" was a major attraction.
Furthermore, since multiple people speak during meetings and discussions, it was important to be able to identify "who spoke."

The main challenge in developing the service was designing a user-friendly interface that would be directly relevant to the actual operations of local governments.
It wasn't just the accuracy of the speech recognition itself that mattered; we also had to consider the entire workflow, including the format in which to send the audio, when to check the transcription results, and how to correct and reuse them.
The AmiVoice API had clear specifications and stable behavior, which was extremely helpful as it allowed us to focus on designing the application itself.

Introduction Results

Comments include: "It sounds natural in Japanese" and "Correction time has been significantly reduced."

Even when compared with globally renowned speech recognition services, AmiVoice delivers outstanding Japanese language recognition accuracy.
By leveraging the Word Registration API, municipality-specific terminology can be registered in advance, enabling high-quality transcriptions without the assumption of manual correction. Furthermore, speaker diarization makes it possible to organize content by speaker, greatly streamlining the creation of meeting records.
Users have expressed strong satisfaction, citing significant reductions in editing time and the production of naturally flowing Japanese text.

The well-organized API specifications allowed for smooth implementation, which was a major advantage as it enabled us to focus on service design.

Future prospects

Plans for secondary use of transcription results and expansion to integrate with other systems.

Going forward, eRex is planned to expand in a direction that will lead to increased efficiency in municipal operations as a whole, including secondary use of transcription results and integration with other business systems. Furthermore, it is expected that its use in the public sector will expand even further if know-how for large-scale processing operations and parameter adjustments tailored to business use are provided.
With these prospects in mind, SHIFT PLUS inc. continues to place its hopes on AmiVoice API as the core Japanese speech recognition engine for eRex.