Project Sekai (プロセカ) - Lost and Found (ロストエンファウンド) - Minori Hanasato AI (RVC V2)

แชร์
ฝัง
  • เผยแพร่เมื่อ 24 ธ.ค. 2024

ความคิดเห็น • 6

  • @sxndypzofficial
    @sxndypzofficial ปีที่แล้ว

    did you make the model itself or

    • @JamesMainChannel
      @JamesMainChannel  ปีที่แล้ว

      Made the model itself

    • @JamesMainChannel
      @JamesMainChannel  ปีที่แล้ว

      drive.google.com/file/d/1pFN_WRkQeH1fc7mTMMjsg03dmU1g6ngS/view?usp=drive_link

    • @sxndypzofficial
      @sxndypzofficial ปีที่แล้ว

      @@JamesMainChannel how many epochs have you trained? and how many minutes of audio data did you use to train it?

    • @JamesMainChannel
      @JamesMainChannel  ปีที่แล้ว

      @@sxndypzofficial unsure, probably 150 epochs and idk what minutes

    • @sxndypzofficial
      @sxndypzofficial ปีที่แล้ว

      @@JamesMainChannel the model is ok, but it needs some improvements
      so here's some of my tips:
      to make a good model, you need at least a 13-15 minute (or more) dataset and you need to train to at least 300-600 epochs
      the smaller the dataset, the higher amount of epochs you need to train
      use harvest for feature extraction if you have a noisy, low quality dataset, use mangio-crepe (hop 64) if you have a studio clear quality dataset
      and watch out for the overtraining