Raine (1/3)
Work-in-progress of my AI Vtuber, Raine. I currently only have the tts and live2d model + vtube studio integrated and implemented, so I've attached a demo of both working together (the lipsync was utter pain to implement)
github.com/Icemaster-Eric/Raine
Raine TTS V2
After spending countless hours trying to get something that doesn't sound completely robotic, I'm proud to announce that I've achieved a tts with configurable emotion, speaking style, and extremely natural speech with a 2x real time factor.
github.com/Icemaster-Eric/Raine
MangaSeer (3/4)
Finally finished MangaSeer's internals, it's now fully working and all of the features have been implemented. Still pretty rough around the edges, but it should be usable now :hu-happy:github.com/Icemaster-Eric/MangaSeer
Waifumem Update
Improved my conversational rag memory thingy, it's now working slightly better and all that
github.com/Icemaster-Eric/WaifuMem
Raine's Vtuber Model (1/3)
I have next to no art experience, so this has definitely been a really huge undertaking for me. The art's still a work-in-progress, but I feel like I might've gotten slightly better at drawing cute anime girls :3
github.com/Icemaster-Eric/Raine-Vtuber-Model
Raine TTS
A pretty simple tts system, the main focus being to apply rvc over any tts output in order to get a consistent voice and high quality generations. Attached an audio file of the rvc model's output.
github.com/Icemaster-Eric/WaifuMem (same gh repo because I couldn't be bothered to make a new one since I also need the tts in waifumem)
WaifuMem
A pretty simple memory retrieval method using vector embeddings and all that jazz, with a focus on conversational memory.
github.com/Icemaster-Eric/WaifuMem
PyroUI (1/4)
This is just the initial setup for this ui component library. It uses web components, and should be fully compatible with any existing codebase for easy integration. The plan is to make it highly customizable in the future and build it into my own ui framework.
github.com/Icemaster-Eric/PyroUI
Just finished designing my Hu Tao PCB Hacker Card! I definitely spent way too much time designing it, but I feel like it was worth it in the end. Can't wait for them to arrive in the mail soon :sip_kanna:github.com/hackclub/OnBoard/pull/783
MangaSeer (2.5/4)
Small feature update this time since the rules have been changed to only allow 25 hours / scrapbook post. I've been spending way too much time on MangaSeer's dictionary, and finally got it to a working stage. It's honestly been quite the struggle, from parsing a huge 200mb Japanese dictionary to squishing around my ui until it was passable.
github.com/Icemaster-Eric/MangaSeer
MangaSeer (2/4)
Finally finished the OCR & popup generation, and added a simple text-to-speech feature for preliminary testing of MangaSeer. Honestly, it's so satisfying to see actual results after many hours of hard work.
github.com/Icemaster-Eric/MangaSeer
Japanese Manga OCR Bounding Box Detection (MangaSeer 1/4)
Spent about 26 hours working on the initial setup. I settled for PySide6 (Qt for Python) to display the screen overlay, and hooked into the internal Windows api for fast screenshots of the active window. Bounding box detection is done using YOLOv10 L (trained on over 500 images painstakingly labelled by yours truly), on the selected region of the active window. Right now everything is quite rough around the edges, but I plan on revamping the ui sooner or later. I'll go into more detail about the project itself once it's ready to ship, but I'm pretty happy to have hit a big milestone so early on.