We spoke to Tomorrow Net, the provider of "CAT.AI," a communication tool that combines chatbots, voicebots, and various management functions, about the reasons for adopting AmiVoice API Private and the effects it has had.

Service Overview

AI customer agents with voicebots and chatbots

"CAT.AI" is a "navigation-type, next-generation AI communication tool" that can be used in multiple modes: voicebot (voice dialogue AI) and chatbot (text dialogue AI). For example, you can use the chatbot while on the phone, and convey information to users visually, rather than just through the "voice-only" guidance of conventional voicebots.

Furthermore, by using advanced AI to memorize conversations, it is possible to have natural conversations as if talking to a person. It comes standard with a wide range of functions, such as speech speed adjustment and image recognition, and can achieve a high self-close rate (AI response completion rate). It is being introduced not only in contact centers but also in a wide range of industries that require communication with customers.

＞Click here for service details

Key Drivers of the Adoption

Switched from other companies due to high recognition of alphabets and numbers

We were originally using Google's speech recognition API, Speech-to-Text. One of our client's key requirements was to "recognize alphabets and numbers." While Google Speech-to-Text is easy to use for everyday conversations and general phrases, we struggled with many misrecognitions when it came to pinpointing and narrowing down the conditions for speech recognition like this. Then, we were introduced to the rule grammar engine that can be used with "Amivoice API Private," which allows you to specify "only recognize alphabets and numbers," so we decided to give it a try.

Introduction Results

We were able to provide functions that satisfied our clients.

The API itself was provided using a standard interface, and the web documentation was comprehensive, so it was easy to implement. The grammar file, which sets the rule grammar, seems easy to understand if you have some experience with programming languages.
However, creating rules that would meet the client's requirements was not easy, as it is with any service or function. At that time, we consulted with AmiVoice Cloud Platform's technical support, who responded quickly and courteously, and we were able to successfully provide the client with a level of alphabet and number recognition accuracy that was satisfactory.

Future use

A name and address engine has also been introduced. Expectations are high for future use.

Our company mainly uses Google Speech-to-Text for general-purpose speech, but as mentioned earlier, we feel that it is difficult to ensure pinpoint accuracy in specialized fields such as "name" and "address" without a dedicated recognition engine.
In addition to the high accuracy of AmiVoice API Private's recognition of names and addresses, the fact that it has a dedicated alphanumeric engine and is easy to integrate were also deciding factors in its adoption.

I thought that cloud speech recognition services generally didn't allow for control over recognition accuracy or rules, so a service like this Rule Grammar Engine was groundbreaking. I look forward to seeing more innovative technologies and services for Japanese speech recognition engines in the future.

*The service names of Tomorrow Net Co., Ltd. displayed and listed on this page are registered trademarks of Tomorrow Net.