issues
search
supikiti
/
Awesome-tts-and-vc
Here is a summary of the conference papers we read.
7
stars
0
forks
source link
An Effective Style Token Weight Control Technique for End-to-End Emotional Speech Synthesis
#7
Open
supikiti
opened
4 years ago
supikiti
commented
4 years ago
リンク
https://ieeexplore.ieee.org/document/8778667
どんなもの?
GSTを用いた高品質な感情音声合成を提案
先行研究と比べてどこがすごい?
話者の特定の感情を表現するための潜在表現を抽出する際,学習するための音声データを選択することが難しい
学習済みGSTモデルにより出力された各感情のembeddingの重み付け和により各音声を表現
技術と手法のキモはどこ?
上に同じ
どうやって有効だと検証した?
教師とする音声を選択するモデルとの主観評価を実施 & 性能向上を確認
議論はある?
次に読むべき論文
Style Token
リンク
https://ieeexplore.ieee.org/document/8778667
どんなもの?
先行研究と比べてどこがすごい?
技術と手法のキモはどこ?
どうやって有効だと検証した?
議論はある?
次に読むべき論文