Categorisation of Reading Ambiguity Type and Building a Dataset of Reading Estimation Tasks for Text-To-Speech Technology
- Project Leader
- SHINNOU Hiroyuki (Ibaraki University)
- Project Period
- October 2022 -
Summary
In recent years, the demand for text-to-speech technology used in applications such as AI Assist has increased.
One of the technological challenges is resolving the reading ambiguity.
There are several types of reading ambiguity, such as the ambiguity with phonetic reading, ambiguity of numerical reading with particles, and the ambiguity with heteronyms.
In this study, words with reading ambiguity are collected and categorised from dictionaries and corpora.
A dataset will be built for the task of reading estimation for words with context-dependent reading ambiguity.
Additionally, a reading estimation system will be constructed, the baseline scores for the dataset will be presented, and the challenges of using this technology will be discussed.
Finally, a text-to-speech system will be built by creating a conversion system from normal text to text for reading out loud.
Furthermore, this system will be able to automatically add ruby to Japanese Kanji characters. Therefore, it can be used as a Japanese language education tool.
