sapokr

Beyond just resampler/wavtool, we’re also preparing to release a voice bank for the character “Kokuno Tetsu” (Organum’s mascot) to make it even more accessible and approachable!

We’re currently working on optimizing the calculations (while maintaining quality) along with a multilingual README! It’s quite a task, but it’s worth it for our users 👀👀

We’re working on SIMD support! It’s significantly faster than the GPU-accelerated (experiments) part! To make GPU acceleration faster, we might have to rewrite the WORLD vocoder part to do GPU calculations 😅

GPU acceleration is experimental, but available with CPU+GPU builds! Most are comparable to CPU speeds, but some are even faster! You can fine-tune it via config! 🥰🥰

GPU acceleration has been added! However, it’s not yet ready for prime time, so it’s not particularly fast, except for a short duration. But we’ll work on improving it 😎😎

We’re currently working on GPU support to improve processing! Initially, we’ll only support partial processing, but we’ll work on accelerating it in the future! 😋

864KB -> 197KB

Effectively reduced cache usage!
Internally, either Quantized or DeltaQuantized is selected and stored!
This results in faster inference! 😁

This is our first release! We’ve created a resampler, wavtool, that can be used with utau. In the future, we hope to make inference even faster and improve the sound quality!

The first release will be released soon…