Mangio-RVCを使ったAIボイストレーニングで、川島れいじのボイスモデルを生の台詞からトレーニングした。前半は、各俳優の実際の演技をもとにしたそれぞれの音声モデルです。後半は、演技を入れ替え、演技者のピッチの違いを補正するためにピッチ・オフセットを調整しました。 その結果が、ここで聴いているものです。 Using AI voice training via Mangio-RVC, I trained from raw isolated dialog, Reiji Kawashima's voice model. I used an already-made voice model for Kenjiro Tsuda for the other half of this experiment. I found that the models sound most like their original actors if driven by the original performance. The first half is each respective voice model, driven by that actor's actual performance. I used a pitch offset of 0 for this half as no change in pitch was required to match the performance to the voice model. For the second half, I swapped the performances and adjusted the pitch offset to compensate for the difference in pitch of the performer. The result is what you hear here. ... https://www.youtube.com/watch?v=k7ZWtQSbKCw