CleanTake


CleanTake — Free Download. Audio repair

CleanTake is a desktop app for repairing dialogue in audio or video recordings when multiple simultaneous captures of the same take exist. Import a primary recording and up to three backup recordings, align their clocks, and review suggested replacements at matching timeline positions. Each repair remains pending until accepted. The repair inspector shows the saved source, original filename, project time range, and corresponding recording time range next to the acceptance control. Unsaved changes are identified separately. Original files remain untouched, and undo and redo are supported. Export options include audio, aligned stems, a source map, and an editable Reaper project that retains repair decisions. Processing runs locally without a separate Python, FFmpeg, browser, model, or GPU computing driver. Desktop downloads for macOS, Windows, and supported Ubuntu systems include processing tools and an offline sample with a deliberately introduced gap in licensed recordings.

★★★★★
5.0(1 ratings)
File size: 180 MB
The latest version of CleanTake is: 0.2.2
Operating system: Windows, Mac OS, Linux
Languages: English
Price: $0.00 USD (Open Source (MIT))
  • Main plus backup import. CleanTake accepts two to four simultaneous recordings of the same take. The primary recording is the repair target, while backups serve as donor sources. Import supports audio and video files, with stream and channel selection. Original files are never modified; the application reads them and stores edit decisions separately.
  • Clock alignment with drift and polarity. The alignment step estimates offset, drift, and polarity between recordings. Uncertainty values are displayed for inspection. When automatic alignment is not reliable, a manual clock can be set. Aligned sources are required before a backup can become an accepted replacement.
  • Shared timeline waveform review. CleanTake shows actual waveforms from the primary and backup recordings at a shared time position. This allows direct visual comparison of the same moment across different captures. The interface supports zooming and moving through the timeline to locate damage and evaluate donor material.
  • Repair proposal inspector. Each suggested repair shows the saved source identity, original filename, project time range, and corresponding recording time range. The acceptance control sits beside this information. A proposal is not applied until accepted. Unsaved changes are marked separately from saved repairs.
  • Bounds, gain, and crossfade editing. Accepted repairs can be adjusted. The user can edit repair bounds, change gain, and control crossfade length and shape. These controls address transitions between the primary recording and the donor source, reducing abrupt changes in tone or room sound.
  • Manual repair creation. When automatic detection misses damage, a manual repair can be created. The user selects a time range and a donor source, then applies the same bounds, gain, and crossfade controls. Manual repair is a core part of the workflow for difficult material.
  • Accept, reject, undo, and redo. Repair decisions are reversible. Each proposal can be accepted or rejected. The undo and redo system tracks changes through the editing session. Rejected proposals remain visible for later reconsideration.
  • Transcript context import. CleanTake can import transcript context with preserved word timestamps. Estimated timing is marked explicitly. This helps navigate to specific dialogue passages and identify locations where recorded audio may contain damage.
  • Source map export. The source map records identities, hashes, clock transforms, and both contributors to every crossfade. This documents exactly which recording supplied audio for each repaired interval. The source map supports verification and archival workflows.
  • Reaper project export. CleanTake can export an editable Reaper session that retains repair decisions. Volume envelopes for source choices remain editable in Reaper. This allows further mixing and finishing work in a separate digital audio workstation.
  • WAV and FLAC audio export. The final repaired timeline can be exported as WAV or FLAC. The output timeline is 48 kHz mono. Optional finishing writes a separate loudness-normalized file while preserving the unmastered mix.
  • Aligned stems export. CleanTake can export aligned stems from the contributing recordings. Stems preserve the original source material after clock alignment. This supports downstream processing and comparison outside CleanTake.
  • Portable project files. Projects can be saved and reopened on another machine. Repair decisions, alignment data, and source references are stored in the project file. Original media files are referenced by path and hash.
  • Local processing without external dependencies. The desktop edition includes its own FFmpeg build and processing tools. No separate Python installation, browser, machine learning model, or GPU driver is required. Processing runs on the local machine.
  • Offline sample project. The first launch opens a workspace with a sample named Sample · recover a missing half-second. The sample contains a deliberately introduced gap in licensed recordings. It works offline and demonstrates the comparison, acceptance, and export workflow.
  • Unsaved change tracking. CleanTake identifies unsaved changes separately from saved repairs. This distinction prevents accidental loss of editing work. The save state is visible during the session.

CleanTake development began as a response to practical dialogue repair problems in production audio. The application was created by Pavan Gupta and contributors. The first public release appeared in 2024. The desktop application is written in Python with a web-based interface layer, bundled with FFmpeg for media processing. Development focuses on local processing, transparent repair decisions, and verifiable source documentation. Releases are provided for macOS, Windows, and supported Ubuntu systems. The macOS builds are ad-hoc signed and not notarized; the Windows installers are unsigned. Platform-specific security warnings may appear on first launch.

Latest Programs