Repository navigation
OpenStudio 0.1.09
What's new
OpenStudio Support lets you inspect an audio or MIDI export and collect existing crash and recovery evidence after closing the DAW. Support opens only when you launch it. You review the archive and decide whether to share it.
You can audition tempo suggestions for audio and MIDI, edit linked crossfades, and enter mixer levels precisely. OGG export uses the built-in Vorbis encoder. Detached pitch editors now wait until their canvas is ready when reopened.
Separate Stems can use less memory on long recordings. Windows and Linux offer to check installed stem tools after an update. Choose Open DAW to skip maintenance, or Check and repair to run it after every DAW session has closed.
Highlights
- Add a separately launched OpenStudio Support application for Windows, macOS
and Linux. Users can collect existing diagnostic files, inspect audio/MIDI
exports and create a support archive without installing Python or using a
terminal. Collection and analysis require OpenStudio to be closed. Nothing
is uploaded automatically; users review and share the resulting archive. - Expand offline evidence with plugin scan logs, filtered plugin/audio settings,
selected OS crash reports and exact native debug identities. Keep incident IDs
separate from the installed version, and preserve matching native symbols and
frontend maps outside installers at build time. - Support preserves plugin scan failure reasons when hiding a path and finds
diagnostic files in relocated Linux folders. After the DAW closes, it reads
existing recording recovery metadata for write, flush and seek failures and
dropped samples. Users can review this evidence before sharing it. - Add audio and MIDI tempo analysis with candidate audition and Undo, editable
linked crossfades, exact mixer level entry and a resizable single-row mixer. - Add optional project-rate copies of imported audio while retaining original
source files, and show conversion progress and saved MIDI-device availability. - Separate Stems can use a supported GPU or run on CPU. On supported Windows
and Linux configurations it can use temporary disk space to separate long
recordings with less RAM. After an app update, Windows and Linux offer to check
installed stem tools before opening the DAW. Open DAW skips the offer for
that version. Check and repair starts maintenance once every DAW session
has closed and stages a replacement when needed. Activation requires validation on the requested
backend. Downloaded models and installed shared generation libraries are kept.
Users can also release an idle generation worker. - ACE-Step, Stable Audio and MiniMax plan model placement around available
memory. Device-specific generation and acceleration options are available
for the configurations described in the AI qualification report.
Fixes
-
Address failures in Separate Stems when the installed runtime cannot run the
model on the selected device. Automatic mode uses CPU for the unsupported
DirectML path and can retry a GPU out-of-memory failure once after the first
worker exits. A successful import keeps the clip's trim and position and can
be undone as one edit. Incomplete results are rejected before adding tracks. -
A reopened detached pitch editor acknowledges readiness after its canvas is
rendered with usable geometry. A slower mount or a hidden view becoming visible
no longer loses the one-time readiness handshake after two animation frames. -
Preserve exact unity coefficients for neutral EQ bands, avoiding small
recursive rounding residuals on platforms that use fused floating-point math. -
Encode OGG with the built-in Vorbis codec, preserving quality 0–10, instead of
requiring a system FFmpeg build with the optional Vorbis encoder. Final
sample-rate conversion retains the existing FFmpeg path. Encoding remains
offline and cancellable, and a failed export cannot replace an existing file. -
Compile out ARA callback timing, logging arguments and per-block debug-state
updates in ordinary builds. Remove periodic heartbeat/fault polling; flush
failure counts at every stop/shutdown, including faults from successive short
takes. An empty drain writes nothing. The Windows crash helper waits on events during
normal use, while crash capture and committed-final-Quit protection remain. -
Reduce repeated work for tiny audio blocks and prepare recording storage
before capture starts. The recording writer publishes prepared FIFO data
without waking its disk thread from each audio callback. These changes do
not establish a universal glitch-free minimum buffer size. -
Prepare native NAM input/track effects for the actual device block size,
reducing unnecessary cabinet-convolution work at small buffers while keeping
the existing preparation reserve for other effects. This adds no cabinet
latency; the strict 8-sample contention target still fails. -
Limit timeline waveform and canvas work to the visible region, keep recording
previews updating during navigation, and reduce ruler label crowding without
changing the snap grid. Mixer resizing releases keyboard focus correctly. -
Preserve joined MIDI note identity, mute state and note releases; export
selected MIDI clips consistently from the context menu and action catalog.
Single-clip export is labelled separately. Correct selection near clip edges. -
Preserve source geometry and fades through project restoration and rendering,
validate final project-rate conversions, and avoid conversion staging names
that exceed Windows path limits. The fader floor produces digital silence. -
Compile callback timing histograms, stage sampling and paired trace storage
out of ordinary builds. Streaming playback also skips diagnostic cache-hit,
read, fallback and source-binding counts while mixing or recording. Missing measurements are reported as unavailable,
including in qualification reports, rather than as zero failures. MIDI event
placement, processing deadlines, meters and existing fault handling remain.
Release packaging rejects an instrumented or mismatched build policy. -
Make the Windows Support window scale its text when moving between displays
and keep its controls reachable with scrollbars on small or highly scaled
screens. These changes run only inside the explicitly opened companion.
Known Issues
-
Ubuntu App Center availability depends on Canonical's classic-confinement
decision and Store processing. Installed Snap checks passed during release
preparation. Snap installations receive updates through the Store. OpenStudio
Support remains a separate launcher. -
The strict Audient ASIO 48 kHz / 8-sample target still fails. After the NAM
preparation/insertion correction, a four-track probe with Dota 2 running
recorded 3,658 late callbacks out of 180,000 and also missed worker deadlines.
The saved previous build recorded 8,948 misses in another run; game activity
was not a controlled replay, so this is not a universal improvement claim.
Raw synthetic-input recording integrity passed, but continuous monitoring,
physical driver loss and subjective quality are not established. The separate
31-track hardware gate also remains failed. Short 64-sample and FX-bypassed
controls do not qualify the loaded 8-sample workload or long sessions.
An explicitly approved trace found graphics-driver interruptions exceeding
the buffer budget and paging during one late callback. These do not explain
every miss, prove RAM exhaustion or establish a past unobserved glitch's cause.
On 5 November the user accepted this as a non-blocking known limitation for
merging PR #36. The measured failure is retained; no glitch-free eight-sample
claim or relaxed test threshold follows from that decision. -
Offline support cannot reconstruct unmeasured CPU scheduling or RAM pressure
at a past glitch, prove that every intended instrument was audible, or replace
a physical signal reference. Redaction is best effort; inspect archives before
sharing. Crash dumps are optional and can contain private information. -
Package checks use unpublished Windows, Linux and macOS artifacts. Windows
App Certification Kit has passed overall and all 13 required tests; two
optional findings remain in its report. This is not Microsoft Store approval.
Mac candidate signing is ad-hoc: Developer ID/notarization credentials are
unavailable, and Gatekeeper acceptance is not asserted. Results apply only to
the exact artifacts recorded in the
candidate qualification record and linked PR checks;
earlier passes do not qualify subsequently changed binaries. -
Automated render and MIDI checks establish objective behavior for their test
inputs. They do not establish subjective sound quality or compatibility with
every external plugin, interface or driver.
Upgrade Notes
-
After an update, users with installed stem tools on Windows or Linux can choose
Open DAW immediately. To revisit repair, choose AI Tools Setup → Repair Stem Separation.
Save your work, close every OpenStudio window and reopen the app from its
launcher. Choose Check and repair, Rebuild or Open DAW. Recording
sessions and maintenance are mutually exclusive. Normal sessions never start
an automatic repair, and Support opens only on request. The CPU-only setup
preference is preserved. A failed candidate leaves the previous runtime active.
Temporary-disk modes need spare disk space, and optional reduced-precision
settings can change the generated audio. -
Use the symbol archive matching the exact app build when investigating an
incident. Automatic timed dumps for running-session UI hangs are disabled in
ordinary builds; explicitly selected OS evidence or a diagnostic reproduction
is needed for those cases. -
Back up projects before opening them in 0.1.09. Existing project
files remain supported; save a separate copy when testing new editing flows.
The 0.1.08 plugin Undo corrections remain included. -
Open Support explicitly from the installed shortcut or launcher after closing
the DAW. No support process, scheduled collector or service needs to run during
recording. Close Support before reopening OpenStudio. -
Keep existing audio interfaces and driver settings for comparison. Use the
buffer/workload combination qualified on your hardware; a successful export
or a larger-buffer test does not qualify an eight-sample recording session.
The application changes since 0.1.08,
application PR #36 and
release PR #37 document this release.