We are conducting research into high-quality speech synthesis from documents and data with the goal of implementing automatic audio broadcasting. We have developed a system for automatically synthesizing the speech of the Weather Reports information transmitted from the Japan Meteorological Agency and reading it out synched to the program time.
- Implementation of broadcast-quality synthesized speech
- Audio data that reflects the word order within the text is selected by choosing a template※ which is similar to the structure of the input text. The result is high-quality synthesized speech with natural intonation.
- Precise adjustment to program timescale
- The speech-rate conversion technology gradually adjusts the speed of the synthesized speech so that the read-out speech is made to end in sync with the program time, even if there are changes such as the exceptions of some sentences in the program content or the additions of extra reports at air time.
In the works
We are considering introduction of this automatic speech synthesis system for weather reports as practical radio broadcasting applications and also considering applications of the algorithm of this research to other fields such as traffic reports, etc.
- ※Template:The template includes a number of text patterns for comparison with all of the text to be read out, in segment units in which similar portions are integrated.
Automatic weather report reading system