Hello again, everyone! Another month, another exciting update for Maghni AI's development. We've hit some major milestones, and we're in the home stretch for some of our biggest goals yet. Let's dive into the details!

Variance Model Restructured: Unique Voices, Unique Flow

To better explain all the incredible work he's been doing, please allow Caleb—our backend developer and co-project manager—to introduce some of the exciting versatility and customization possible with Maghni AI's newly restructured variance model!

Hi! I've been working on the timing model a TON lately. It's something I'm extremely proud of, and I hope y'all like it.

One of the major facets I've tried to focus on while working is optimization. The model is EXTREMELY fast, meaning less time waiting between changes and audio rendering in Maghni. The files are small, meaning they can easily bundle with the voice and help save storage.

Additionally, I've tried to make it very feature-rich. In Maghni's global settings and specific note settings, you'll see options for "maximum consonant percentage" — a term that means how much of a note the consonants that start the next note are allowed to take up. It's kinda hard to picture without having it in front of you, sorry, but essence, this setting allows you to tune the naturality of the voice. No matter how high you set the percentage, the timings will never be stretched unnaturally—they will only be shrunk if they would otherwise not fit.

A lower percentage would make smaller notes in fast verses much more audible. With the default setting of 60%, the vowels are guaranteed at least 40% of each note. meaning each syllable will have its time to shine.

Another interesting feature I've implemented recently are phoneme maps. We promised these earlier, but didn't go too far into describing them. By default, Maghni will use a modified but faithful X-SAMPA, similar to VOCALOID or Delta English for UTAU. There are many other options outside of X-SAMPA, however!

By default, we will have many implemented. For languages such as Spanish, whose consonants will almost always correspond to the same phoneme, we will provide a "native"-type phoneme map. The phonemes of the word "bebe" would be written in this phoneme map as [b e b e], while in X-SAMPA they would be written as [b e B e]; despite the two bs being the same in the native-style map, they will correspond to b and B in X-SAMPA respectively.

We have also implemented Romaji for English phonemes, Romaji for Japanese phonemes, Pinyin and Zhuyin for Chinese phonemes, and Romaja for Korean phonemes. These phoneme maps are different from dictionaries; a dictionary will take a native word (such as, for Japanese, "ao," "あお," "アオ," or "青") and will return the phonemes. These phonemes will be displayed differently according to the map, with the base/raw phonemes always in our default X-SAMPA implementation.

We have not limited phoneme map creation to ourselves, either. You will be able to create your own, custom phoneme map using any Unicode character and send it to others online. While we hope to provide as many as possible, we understand everyone will have their own preference for how to use Maghni!

I hope you find the timing model's settings and the phoneme maps easy and fun to use!

Backer Rewards Issues Resolved: Production is on Track

We know many of you have been waiting patiently for updates on physical backer rewards, and we're happy to report that all manufacturing issues have been resolved!

After frustrating delays caused by Hurricane Helene's impact on local businesses in our area, we've finally worked through the setbacks. Production is now running smoothly, and we're moving full steam ahead.

We're incredibly grateful for your patience throughout this process. To ensure a seamless experience, all rewards will soon be shipped to VocaTone for final distribution. We know this has been a long wait, but we promise—it'll be worth it!

Thank You for Your Support!