Repository navigation
Ruach Studio 2.1.0-rc1
Pre-release2.1.0-rc1 is the first candidate for 2.1.0: a pre-release with everything made since 2.0.9 and the fixes of 2.0.10. Pinokio's Update and a git pull of main bring it now, as main is at 2.1.0-rc1; the Studio's own updater offers it when you ask for release candidates (Engine → Updates → «release candidates too»). 2.0.10 stays the latest release until 2.1.0.
The Musician
- The Section editor: Edit › Section editor… opens a take as a strip of its sections over the whole room. Click a section, queue one edit for it (replace it, re-sing a line, sing it once more, insert an interlude or a copy of one of the score's parts after it, repeat, remove, re-sing to the end, extend), then Assemble the take: the queue is sung as one new take, a child of the source, which stays as it is. A take with no score of its own (Direct, or brought in) gets one written from its sound first, and the new take goes into the player when it lands.
- Follow my sections is off by default (it was on in 2.0.10): it sits under the planning modes, says what it does and what it costs, and its line tells how long a plan these lyrics would make. It is for short songs; past 4:00 Generate asks first, with the numbers measured. [Interlude] with words stays an interlude.
- The song lives in the Writer from its first save: Ctrl+S and every Generate of a titled song keep it in its own Writer document, each changed text a version.
- MIDI export in one window: the score's MIDI at once, or MuScriptor on the mix or per stem, on the take's own beats, with a track per instrument.
- Music cache in 8 bits (Engine → Compute, off by default) for smaller cards: about half the cache's memory.
The Narrator
- Three columns: the text on the whole height, «At hand» beside it, the voice and the engine at the right; its readings on the right with reactions, as the Musician's takes.
- The lyrics editor's tools in its text, and «[» for the engine's tags and the voices.
- The first 2, 3 or 5 minutes, and up to four versions at once, each with its own seed.
- A director's line for Drama 3: ✨ lets the Writer's model read the whole text and write one line on how to read it; you edit it, and Drama hears it before every piece. Top-p beside the temperature.
- Your own voice: rename it, clean its clip, give it a male or female profile.
- Every reading kept in the Writer with its versions; 1.0 when no version is set.
The Library and the workspaces
- Workspace emoji: an emoji for each workspace and section, in an even column of the tree; a chooser sorts A to Z or by size and can show the archived or hide the frozen and the empty. Names that began with an emoji are moved into the field once, on the first start.
- History of plays and downloads: Ctrl+H and Ctrl+J, or ☰; 30 days of plays and 90 of downloads.
- Lower, tidier cards; a card's workspaces opened in one click; play-on from 0:00, in the list's own order; one play button style everywhere; ✕ at the top right of every dialog.
- Deleting a take frees all of its files: the Narrator's pieces, the Artist's pictures drawn for it, its marks.
The DAW and the Trainer
- The DAW plays «Live» in the browser or «Forge», rendered by the Studio from the playhead; a lane can be a spectrogram; the detached windows follow your screens.
- The Trainer's cards and clock say a run as it is: both towers' learning rates, the phase's epochs, the clock from the recent steps with their evaluations.
Windows
- RuachStudio.exe starts the Studio with no console windows, keeps the lab and the engine running, and makes a tester's report on your Desktop that you send yourself.
- The Weights panel and the system bar work natively (issue #15): no more «module 'os' has no attribute 'statvfs'»; downloads run through Git for Windows' own bash (never WSL's), and without it the panel says what to install. The models a song needs also come with
.venv\Scripts\python launcher\fetch_models.py. - The first start's install shows its steps and asks about the models: its window stood empty and closed at once, the models' question unanswered; now it shows every step and takes your answer, and an install stopped before its end goes on at the next start.
- With no NVIDIA card the Painter's sizes, the Narrator's cards and its voices on the processor answer instead of erring, and the updater points to the tray's Check for updates.
- The models' download says how far it came in the install window, and the Trainer says training runs on Linux for now instead of sending you to bash scripts.
- The install brings the same packages as on Linux (the LoRA's measure and the score's MIDI needed two more), and a new version unpacked over an older folder brings its packages up to date before the Studio starts.
Pinokio
- Install, Update and More models take your answer in their terminal: «Download …? [Y/n]» can be answered now. Thanks to @arkinou.
What to try
- Open a take in the Section editor, replace its bridge and insert an interlude after it, and assemble them as one take.
- Read the first 3 minutes of a text in the Narrator, then four versions of a paragraph at once.
- Give your workspaces emoji and sort the tree by size.
- Export a take's MIDI per stem and open it in your DAW.
- On Windows: open the Weights panel and tell us what it says.
Install or update
- Linux, from git or Pinokio: Pinokio's Update or
git pullbrings 2.1.0-rc1; in the Studio, Engine → Updates with «release candidates too» ticked offers it; orgit fetch --tags && git checkout v2.1.0-rc1. - Linux, the AppImage: download
RuachStudio-2.1.0-rc1-core-x86_64.AppImagewith its.sha256,sha256sum -cit,chmod +xit and start it in place of the old one; your models, songs and settings stay where they are. - Windows 11: download the Windows zip and
SHA256SUMS.txt, check the sum (certutil -hashfile FILE SHA256), unpack into a short folder, startRuachStudio.exe. - macOS 13.3 or later, Apple Silicon: download the Mac zip, check its sum (
shasum -a 256 FILE), unpack it and openRuachStudio.command.
Known limits
- A release candidate: what is in it works and was checked; the last polish comes in 2.1.0.
- A long song can end before its planned Outro: on the 3B models two plans past 5 minutes, followed section by section, stopped 44 and 57 seconds before their plan's end with the Outro unsung. «Follow my sections» asks before a plan past 4:00; for a long song, split the lyrics or extend the take.
- Windows and the Mac are still a beta: not every room is there yet, and nothing is signed.
- AppImageUpdate follows the latest release (2.0.10), not candidates; update an rc AppImage by downloading the next one.
- Linux needs an NVIDIA card with the driver 570 or newer for full speed; an Intel Arc runs the engine as a beta.
Downloads
- The source tarball and the tag
v2.1.0-rc1. RuachStudio-2.1.0-rc1-core-x86_64.AppImagewith its.sha256and.zsync.- The Windows 11 zip and the Mac zip (beta), with
SHA256SUMS.txt.
Thanks
To uboneh-lab for issue #15, found and worked around by hand before we knew; to cynic2012 for issue #13; to Timon (timoncool) for yue2.cpp's fixes and the Windows build recipe; to m-a-p for YuE2, ServeurpersoCom for yue2.cpp and ggml-org for ggml; to Fish Audio for Drama 3 (Built with Fish Audio) and to Boson AI for Higgs TTS 3 (Built with Higgs TTS 3 licensed from Boson AI USA, Inc.).
Everything, entry by entry: CHANGELOG.md and heresy/docs/HERESY.md.