Airion Inc., a University of Tokyo-born venture specializing in AI implementation for the manufacturing industry, has developed a voice synthesis AI for "Tommy-kun," a partner robot developed by Toyota Motor Corporation's Frontier Research Center.
■ What Is "Tommy-kun"?
"Tommy-kun" is a communication robot developed by the Frontier Research Center, a research division within Toyota Motor Corporation. Tommy-kun can communicate with people through gestures such as body and hand movements, changing facial expressions on the display that serves as its eyes, and spoken conversation.
For this project, Airion developed the voice synthesis AI to be implemented in Tommy-kun. Specifically, we built a system that recognizes speech directed at Tommy-kun, has a natural language processing AI compose a response, and then delivers that response through voice synthesis AI in the voice of newscaster Yuta Tomikawa (of Toyota Motor Corporation).
■ What Airion Developed
Given that this voice synthesis AI is intended for "conversation modeled on announcer Tomikawa's voice," we focused our development on the following three points.
1. Support for specialized terminology
Going forward, Tommy-kun is expected to hold conversations related to Toyota Motor Corporation, such as guiding visitors at the Toyota Kaikan. This requires accurately pronouncing uncommon specialized terms — proper nouns such as "Toyota Motor Corporation," "Frontier Research Center," and "Tommy-kun."
To address this, we equipped the system with a feature that allows new specialized terms, along with their readings and accents, to be registered from outside the system, enabling Tommy-kun to explain content that includes such terminology.
2. Improved intonation accuracy
Thanks to recent advances in AI technology, synthesizing speech in the voice of a specific person is becoming increasingly accessible, but correctly controlling intonation is still very much a developing area. Particularly when specialized terms are involved, the intonation of each word must be specified correctly and reproduced faithfully by the AI.
For this project, with the cooperation of newscaster Tomikawa himself, we experimented extensively with the training corpus, recording environment, and speaking style to create high-quality training data and improve intonation accuracy.
3. Faster conversational response
One of the most important elements of natural spoken dialogue is response speed. If more than one second passes between someone speaking to Tommy-kun and the start of its reply, the conversation feels noticeably slow.
To mitigate this, we selected a voice synthesis model with excellent inference speed, shortening the time between when a person speaks and when the robot responds.
■ Looking Ahead
Beyond voice synthesis, Airion develops a wide range of AI solutions for the manufacturing industry.
We implement AI across every function — product development, design, procurement, manufacturing, and more. If you are thinking, "We want to develop new products using AI" or "We want to streamline existing operations with AI," please feel free to contact us through our website.
Tommy-kun is scheduled to be exhibited at the Toyota Kaikan (Toyota City, Aichi Prefecture) in the future.
Toyota Kaikan (Toyota City, Aichi Prefecture)
