• Search Research Projects
  • Search Researchers
  • How to Use
  1. Back to project page

2022 Fiscal Year Final Research Report

Can AI Rakugoka entertain people? -Improved expressiveness of rakugo speech synthesis and automatic generation of storytelling

Research Project

  • PDF
Project/Area Number 21K19808
Research Category

Grant-in-Aid for Challenging Research (Exploratory)

Allocation TypeMulti-year Fund
Review Section Medium-sized Section 61:Human informatics and related fields
Research InstitutionNational Institute of Informatics

Principal Investigator

Yamagishi Junichi  国立情報学研究所, コンテンツ科学研究系, 教授 (70709352)

Co-Investigator(Kenkyū-buntansha) Cooper Erica  国立情報学研究所, コンテンツ科学研究系, 特任助教 (30843156)
Project Period (FY) 2021-07-09 – 2023-03-31
Keywords音声合成 / 落語 / 深層学習 / 言語生成
Outline of Final Research Achievements

We have conducted machine learning research to construct a DNN-based rakugo performer’s speech synthesis model, which can generate natural-sounding audio that entertains listeners by performing rakugo like a professional performer. First, we constructed speech synthesis models called Tacotron, Transformer, VITS, and FastPitch on our rakugo database. We also developed an explicit modeling method for nonverbal information such as laughter, which is frequently used in rakugo, and proposed a new method that uses the approximate shape of speech waveforms as input units. Furthermore, since it is impossible to entertain listeners if rakugo stories are exactly the same every time, we also studied a framework for automatic generation of rakugo stories using neural language models such as GPT-2, BART, and T5.

Free Research Field

音声情報処理

Academic Significance and Societal Importance of the Research Achievements

伝統話芸である落語を深層学習で再現し、AI噺家を実現しようと言う、本研究の試み自体が、情報伝達や質問回答を目的とする従来の音声対話システムとは目的が全く異なり、ユニークでかつ学術的意義のある試みである。構築された音声合成システムの比較実験からは、AI噺家が人を楽しませるためには、従来の音声合成の自然性に関する評価指標のみでは解決できない事も判明し、音声合成のモデリングのみならず評価体系を抜本的に変化させる必要があることも判明した。また同時に、Tacotron、 Transformer、FastPitchという種々のEnd-to-end音声合成モデルの中でどれが落語音声に適しているかも判明した。

URL: 

Published: 2024-01-30  

Information User Guide FAQ News Terms of Use Attribution of KAKENHI

Powered by NII kakenhi