AI announcer has made a debut on “News Check 11” program

The AI announcer “News Yomiko”, which uses text-to-speech synthesis technology developed at STRL, has made a debut on a news program “News Check 11” on the NHK General TV channel. The text-to-speech synthesis is implemented with “deep learning”, a kind of AI technology. The voice in which the announcer reads the news has a tempo that is smoother and easier to understand than a normal human voice. To achieve such a voice, the AI was trained on contextual voice features using a large volume of news scripts and announcer speech. Using the speech of announcers, which embodies announcing expertise, as training data results in a synthesized voice that has a natural news tone. We also plan further training with more broadcasts.

In future work, we will continue to move forward with R&D to achieve text-to-speech synthesis that is even closer to the human voice.