Ruach Studio 2.0.8: the all-in-one release

[igrbible]

Ruach Studio 2.0.8 is out: the release. We cut it in honour of the whole Studio at once, two days after 2.0.7: every room has changed.

The full article is in the Wiki: Ruach Studio 2.0.8: the all-in-one release. Here is the short of it.

• The Writer in two worlds, the Musician's and the Narrator's, each with its own tasks, templates, charter and preferences. The Writing room writes into the open document and says which, never changes your lyrics unasked, keeps every answer with its brief and model, counts the question before it is sent and stops when you say so.
• The Narrator grows. Boson AI's Higgs TTS 3 reads beside Drama 3 as its second engine, fetched from Boson's own repositories once you accept its licences, never mirrored by us. Figures are said in words in the case the sentence asks (Russian, Ukrainian, Greek); Russian (A. A. Zaliznyak's Grammatical Dictionary, from gramdict's data) and Belarusian (GrammarDB) stresses come with it beside Ukrainian's; polytonic Greek is read as monotonic; a seam guard and a de-clicker take out the knock where pieces meet; a voice builder makes voices from a description, and twelve stock voices made by Drama 3 itself come with the Studio. Built with Fish Audio.
• The Videographer's two viewers, as an editor's: the animatic or the clip on the left, the edit's program on the right, and the edit's own tracks for the shots, the titles, the karaoke, the look and the song, in a font of your own.
• Every model kept at home by its content, and every pin asked on Hugging Face before each release: a loss stops the cut.
• Long songs whole by default. A song the model would sing past one window of its context (about 6:16 to 7:52 on the whole context, shorter on a card with less memory) now ends there, whole, with a toast; Sampling › Sound and output › Long songs › Pieces · experimental sings it to its end in pieces, as 2.0.7 did.
• The engine measured on Vulkan beside SYCL on the Intel Arc: the same sound, 1.6 to 3.4 times slower, so the Arc stays on SYCL.
• The UIX pass, under Viktor's law «Максимум компактности, минимум текстового кнопочного клаттера»: 103 findings in every room, all 29 of the first weight mended before the cut; one set of icons, one height for a field and one for a bar, the Studio's own picker for long lists.
• One set of install sizes everywhere: the install 22.5 to 25.9 GB, every room 122 to 240 GB. The AppImage says what its first install downloads before anything comes, and waits for the network instead of failing.

To update: git pull or Pinokio's Update, or Engine → Updates; the AppImage is on the release page. Higgs TTS 3 comes from Engine → Models, by your own hand. Every change, line by line: CHANGELOG.

Thanks to Boson AI for Higgs TTS 3, to gramdict and the Belarusian NLP community for the dictionaries, and to Fish Audio and audio.cpp for the Narrator's first voice; the credits as their licences ask are in the article and in README. Questions and findings are welcome right here.