CleanTake 0.2.2

CleanTake 0.2.2

See which microphone supplies a repair before you accept it.

The repair inspector now places the saved recording name and original filename
(when different), project time range and corresponding recording time range beside
Accept repair. The recording range accounts for the source's saved offset and
clock drift. Displayed times round to the nearest millisecond.
Unsaved edits are marked separately, and acceptance stays unavailable until a
saved, aligned source covers the complete passage.

This refinement follows a community suggestion to make the replacement visible at the decision.

The recovery engine, dependencies and project format are unchanged. Existing
projects retain their decisions. The working demo remains a recording of 0.2.1;
its video, GIF and audio files have not been relabeled or replaced.

Download the matching desktop installer for Mac, Windows or supported Ubuntu
systems. The app includes the processing tools and an offline sample. Mac builds
remain ad-hoc signed and unnotarized; Windows installers remain unsigned.
Installation details.

Verified release evidence

• Qualified source: e9c1435328144fcaecf0b211543321174a7cb8e2.
• Regular checks: Checks 34255663566: 380 Python tests in each of five environments, five fresh installed-wheel smoke checks, 33 browser cases across Chromium/Firefox/WebKit, and 11 production-bundle Chromium cases. Python jobs each recorded five scoped skips and two upstream warnings.
• Native installers: Native desktop 34255663452: all six Mac, Windows and Linux targets passed 13 runtime checks and nine installed-app checks each. Every target retained all 24 project files through uninstall and reinstall.
• Fresh Python package and demo compatibility: Wheel installation outside the checkout, exact studio assets, real recovery/export/archive smoke, unchanged demo compatibility, and a byte-identical wheel rebuilt from the source archive all passed.
• Release assets and corresponding source: All 12 assembled assets matched the new manifest and all 11 checksum entries. The native corresponding-source archive includes the matched dependency sources, recipes, notices and build evidence.

These results describe the named builds and environments. They do not establish
universal installation compatibility, naturally damaged-session performance or
independent listening preference. Existing 0.2.0 and 0.2.1 releases remain available.

CleanTake 0.2.1

CleanTake 0.2.1 fixes estimated transcript timing that could appear exact in Azure phrases, Deepgram paragraphs and grouped AssemblyAI words. The correction covers all supported JSON source layouts and estimate aliases, including words with missing endpoints. Adjacent and overlapping precise passages retain their own labels.

Audio recovery, sample-based edits and dependency versions are unchanged.

Existing projects: the correction applies to transcripts imported with 0.2.1. Saved project and archive entries retain their previous labels. Re-importing adds another transcript; it does not replace the old one. Preserve edited projects and use a separate project for a clean corrected transcript view when needed.

Choose the installer for your operating system and processor. Python, the studio, processing libraries and media tools are bundled. Mac builds are ad-hoc signed and not notarized; Windows installers are unsigned. See the desktop guide for first-open steps.

Build commit: 0730f04. Application checks · Six-target native qualification.

Assets include six installers, the wheel and source archive, the unchanged licensed demo, corresponding native source, a release manifest and SHA-256 checksums. The demo retains its original intentionally injected fault and provenance. This patch adds no natural-fault recovery or listener-preference result.

CleanTake 0.2.0 — desktop apps for Mac, Windows and Linux

CleanTake now opens as a desktop app with its audio tools, processing runtime
and sample recordings included. Install the matching Mac, Windows or Ubuntu
package, open the sample, compare the missing passage against its backup, and
export a repair. No separate Python, FFmpeg, browser or model download is needed.

The full recovery workflow is shared with the command-line edition: align two
to four simultaneous recordings, inspect source changes, make reversible repair
decisions, and export WAV/FLAC, stems, a Reaper project or a portable project.
Every replacement comes from an actual recording.

Choose your installer

• Mac: Apple silicon mac-arm64.dmg or Intel mac-x64.dmg; macOS 14 or newer.
• Windows: Intel/AMD win-x64.exe or Windows on ARM win-arm64.exe.
• Ubuntu desktop: Intel/AMD linux-amd64.deb or ARM linux-arm64.deb.

The Windows installer contains its payload and installs for your user. On Mac,
drag CleanTake from the disk image into Applications. On Ubuntu, use your package
manager to install the Debian package and any required desktop system libraries.
Linux portable archives are not part of this release.

Publisher trust: these Mac applications are ad-hoc signed and not notarized.
Windows installers are unsigned. Operating-system protection may warn or block
first launch; the desktop guide
explains the actual support scope. Bundling dependencies does not establish a
verified publisher. The installers do not disable operating-system protections.

Try the included sample

On first launch, an empty workspace gets Sample · recover a missing half-second.
Open it, select the passage near 17 seconds, compare Original with Source,
then choose Accept repair. Listen in Repair mode and export. The suggestion
starts pending.
Reinstalling or updating the application preserves your saved projects.

The attached cleantake-demo.zip is the same portable sample with its repair
already accepted. Its original public bytes are preserved. Real recordings come
from the AMI Project Consortium under CC BY 4.0; the headset has a labeled injected
half-second gap. Keep the attribution and modification record
with the sample when redistributing it.

Validation

All six targets passed native run 34177076224
at commit 43ad902829714aeb02e9773b65abdb0cbc0b042f: 13 frozen-runtime gates and
9 installed-app gates per target, with all 24 saved files retained through
uninstall and reinstall. Checks include actual audio-clock progression, repair
history, exports, reopening saved work and worker cleanup.

The tested systems were macOS 14.8.9 ARM and 15.7.9 Intel, Windows 11 build 26200
ARM and Windows Server 2025 build 26100 x64, Ubuntu 24.04 ARM and Ubuntu 22.04 x64.
These hosted runners had build tools installed; the actual app ran with developer
tools absent from its search path and its native dependencies inspected. This
is not certification of a pristine physical computer, every OS version or every
audio device. Publisher trust under a final browser download remains separate.

The regular checks
at the same commit passed 340 Python tests with 5 scoped skips on each of five
jobs, all five fresh package installations, 24 browser tests and 8 against the
bundled studio. The attached wheel and unchanged demo also passed a fresh local
installed check, including real recovery, a 48 kHz 960,000-frame export, exact
primary audio outside the accepted repair and archive round trip.

The validation record
explains the exact platforms, skipped checks, portable Linux exclusion and trust
limits. The native source supplement preserves the original reports used by
release assembly.

The final source distribution was safely extracted outside the checkout and
rebuilt to a wheel that is byte-identical to the tested attachment. Its extracted
Electron source helper and all 33 pinned local packaging recipes passed. Release
assembly verified all six installer/report pairs, their component inventories
and the retained Python/demo validation before producing the attached checksums.
The release tag adds documentation to the native build commit named above.

Source, checksums and command-line tools

SHA256SUMS covers the installers and attached artifacts. NATIVE-RELEASE.json
records the exact native build commit, tested targets, report hashes and artifact
hashes. The Python wheel and source distribution remain available for command-line
and browser workflows; those installations still use an external FFmpeg build.

CleanTake 0.2.0 native dependency source
contains the matching FFmpeg 9.0.1 and SoundFile sources for each native target,
plus the Electron 44.2.0 / Chromium 152.0.7977.76 source supplement, patches, notices
and build instructions. Electron's modified FFmpeg is included separately from
the backend's media tools. The installed Electron and Chromium notices are retained.
See the source supplement guide
for exact component scope and rebuild instructions.

Automatic suggestions need listening review and can miss damage. Output is a
48 kHz mono timeline. A word missing from every recording has no recorded
replacement. Independent listening preference, organic-damage coverage and
editor-time advantage remain unestablished; see evaluation.

CleanTake 0.1.0 — recover dialogue from your backup recordings

CleanTake recovers damaged dialogue from simultaneous recordings. This first release includes the local browser studio, CLI, saved projects, reversible repair decisions and portable exports.

Import your main microphone and its backups, align them, compare Original/Repair/Source at the same moment, and keep each source switch editable. Export float WAV or PCM24 FLAC, aligned stems, a Reaper session, a contributor source map or a portable project archive. Optional finishing preserves the unmastered mix and reports actual loudness, including a warning when its target is missed.

Install and try the sample

Install FFmpeg/FFprobe, then install the attached wheel with Python 3.12+:

uv tool install --python 3.12 ./cleantake-0.1.0-py3-none-any.whl
cleantake doctor
cleantake demo
cleantake studio

The wheel includes the studio and licensed offline sample. cleantake-demo.zip is the same sample with its sole automatic proposal explicitly accepted; import it through the studio's portable-project action to inspect that result. The default cleantake demo leaves the proposal pending for your own review.

The demo recordings are from the AMI Project Consortium, meeting ES2004a, and are licensed under CC BY 4.0. The headset includes an injected half-second gap; the accepted repair uses the simultaneous lapel recording. Keep the attribution and modification record with the demo when redistributing it.

Evidence and limits

Native release checks passed: 242 Python tests on Linux (3.12/3.13/3.14) and macOS, and 238 on Windows with four explicitly POSIX-only skips. All five jobs built and installed the wheel outside the checkout, then completed a real sample recovery/export/archive/studio check. All 24 browser workflows passed across Chromium, Firefox and WebKit, followed by eight additional bundled-studio workflows under production security headers. The published evaluation includes untuned AMI excerpts, missed proposals, exact contributor reconstruction, a 30-minute resource run and the actual same-input finishing baseline.

Automatic proposals need listening review and can miss damage. The output timeline is 48 kHz mono. There is no recorded replacement when every source lost a passage. Independent listening preference, organic-damage coverage and editor-time advantage have not been established.

Installation · Editing guide · Evaluation · Licenses