This case features Tokyo System Research Corp., a company that provides AI solutions to support operator training in call centers. We spoke with Mr. Arai, Assistant Manager of the Common Business Solutions Department, about the background behind adopting the AmiVoice API and the results they have achieved.

Background and Challenges in Using Speech Recognition

AI‑Powered App Supporting Operator Training

AI self-training is a service in which an AI robot acts as both a customer and a supervisor, reducing the training burden on call center operators and improving efficiency.
The background to the launch of this service was that in meetings with customers, we learned that many of them had issues with training their call center operators, and that role-playing in particular was taking up too much of the time of trainers. This led us to develop an application that would allow training to be done even in spare time.

In AI self-training, AI analyzes the content of conversations, clarifying matters that need to be communicated and pointing out corrections to inappropriate expressions. In addition, by teaching the AI ​​variations (call scripts) for responses to contract details, product inquiries, and even complaints, training can be tailored to a variety of situations. Another major feature is that training that is close to the actual workplace can be done during the operator's free time, without any time restrictions.

Click here for service details＞

How to use AI self-training

Key Drivers of the Adoption

Chose Our Solution for Superior Accuracy and Real‑Time Performance

In our AI self‑training system, we use speech recognition technology to convert the trainee’s spoken input into text.
To accurately evaluate what the trainee says, high recognition accuracy is essential, and real‑time processing is equally important to create a natural, role‑playing conversation experience.
Initially, we used another company’s speech recognition API, but we struggled with its low accuracy and lack of real‑time performance.
While searching for a solution that could address these issues, we discovered the AmiVoice API. After testing the speech recognition demo available on the website, we found that it offered both high accuracy and real‑time conversion, which led us to adopt it.

AI self-training screen image

Impact of the Implementation and Future Outlook

Reduced Accuracy Complaints, with Sentiment Analysis Coming Next

When we were using a voice recognition engine from another company, we received complaints from customers about the accuracy of the recognition, but since adopting AmiVoice, the number of complaints has decreased significantly. The fact that the results are returned in real time has also been well received.
Furthermore, during development, we were able to immediately use an environment in which we could test the voice recognition API, and sample programs and other information for developers was also readily available, which allowed us to efficiently proceed with testing and development.
Although I did have some trouble connecting Websocket and audio data, I was able to get through it without any issues thanks to technical support, and I am very satisfied. The fee was also very reasonable, allowing me to carry out various tests.

In the future, we are considering using the optional "emotion analysis function" of the AmiVoice API, which can determine emotions from voice, from the perspective of operator training.
Other than AI self-training, if there is a demand for developing services or products that use voice recognition, I would definitely like to use AmiVoice.