README.md

metadata

title: README
emoji: 📚
colorFrom: red
colorTo: yellow
sdk: static
pinned: true

🧪 ShoukanLabs ー召喚Labsー

Shoukan 召喚【しょうかん】 ー can be translated into either Summon or Summoning.

🔎 What is ShoukanLabs?

We are a small (non-company) organisation aiming at producing cutting edge models

• ButterCream
• AshtakaOOf
• Alex
• High Speed Rail Enjoyer
• Neggles
• ZeroCool
• Soshyant
• Wooser
• Karesto

📚 Our Projects

• Future Projects

  • None as of this current time

• Current Projects

  • A Large collection of projects surrounding TTS
    • Vokan-V2 - An iterative improvement on the Vokan TTS model, featuring several architectural improvements.
      • More details soon...
    • VoPho - A universal meta-library for phonemisation under the MIT license, with support for single language and multi-code text! These phonemisers go under an accuracy verification process, to esnure the outputs are sound.
    • VokanPipe - Our dataset curation tool designed to make dataset production simple, efficient, and largely unsupervised.

• Previous Projects

  • AniSpeech - An expressive dataset used to train Vokan V1 (unfortunately, not of the best quality, we're working on it!)
  • Vokan - An expressive StyleTTS2 finetune with better 0-shot capabilities
  • OpenNiji - A finetune aimed at replicating Nijijourney on Stable Diffusion.
  • OpenNiji-V2 - A second finetune made to replicate the Nijijourney style more accurately.

💡Our Philosophy

At ShoukanLabs, we believe in:

• Contributing to the community
• Cutting-edge AI research (and teaching old models new tricks)
• Collaborative development
• Not being limited to a hobbyist level even if we're hobbyist developers