Skip to content

Download and install

Choose a client, then prepare the separate transcription runtime if you want to process audio. Browsing an existing library does not require model weights or a model API key.

Release and current source

The v0.9.0 release provides four desktop installers. GitHub Releases is the source of truth for versions, checksums, and package notes.

The handbook also describes later source features: resizable/collapsible columns, the compact macOS toolbar, spoken-language checkboxes, and summary output-language selection. Build current source for them. A website update does not upgrade an installed app.

Platform v0.9.0 download Contents
macOS / Apple Silicon Swift DMG Native app and CLI wheel; Python, tools, and weights remain separate
macOS / Apple Silicon Qt DMG Qt client/libraries; external CLI for processing
Windows / x64 Qt installer Qt client and VC++ runtime installation; configure CLI separately
Linux / x86_64 Qt AppImage Client; no promoted stable Linux ASR backend
iOS / iPadOS 17+ Source build Reader/subscriptions; no automatic IPA, App Store, or TestFlight distribution

Weights are never bundled. Actions trial artifacts are different from releases and can require GitHub login.

Verify the package

Download SHA256SUMS from the same release. Per-asset .build.json records identify the tag and build. From the download directory:

shasum -a 256 SonaFolioSwift-v0.9.0-macOS-arm64.dmg

Get-FileHash .\SonaFolioQt-v0.9.0-Windows-x64-Setup.exe -Algorithm SHA256

sha256sum SonaFolioQt-v0.9.0-Linux-x86_64.AppImage

Compare the exact platform/file entry. A matching hash verifies the release bytes, not a substitute for reviewing the source and notes.

macOS / Apple Silicon

The native app requires macOS 13+. Stable Chinese ASR and automatic macOS packages target Apple Silicon. Choose Swift for the native workspace or Qt for the cross-platform client.

  1. Open the DMG and copy the app to Applications.
  2. Launch it. Packages use ad-hoc signatures without Apple notarization. For a trusted download blocked by macOS, use its normal Privacy & Security → Open Anyway flow.
  3. Prepare the CLI below and models/FFmpeg, then run doctor.
  4. Check Settings → Overview in Swift. Optional cloud readiness is not needed for local-only work.
  5. In Qt, select the external executable, normally ~/Documents/SpeechToTextData/env/bin/atot.

The compatibility entrypoint Audio2TextCli remains intentional. The included wheel is not Python, FFmpeg, or model weights.

Windows / x64

  1. Run the unsigned Qt installer and follow its dialogs, including Microsoft Visual C++ Redistributable installation.
  2. Prepare Python and CLI; Windows does not need Swift/Xcode.
  3. Supply ffmpeg.exe/ffprobe.exe through PATH or vcpkg.
  4. Prepare the faster-whisper model.
  5. Run the managed atot.exe doctor and select the same executable in Qt.
  6. Supply matching CUDA/cuDNN or explicitly choose CPU.

Defaults are faster-whisper, cuda, and int8_float16. Missing CUDA fails explicitly. Intentional CPU:

$env:ATOT_FASTER_WHISPER_DEVICE = "cpu"
$env:ATOT_FASTER_WHISPER_COMPUTE_TYPE = "int8"

These variables affect processes launched from that environment. A separate desktop shortcut does not inherit another PowerShell session's values. See CUDA troubleshooting.

Linux / x86_64

Use Ubuntu 24.04 or a compatible glibc/desktop base:

chmod +x SonaFolioQt-v0.9.0-Linux-x86_64.AppImage
./SonaFolioQt-v0.9.0-Linux-x86_64.AppImage

Without FUSE:

./SonaFolioQt-v0.9.0-Linux-x86_64.AppImage --appimage-extract-and-run

This is client availability, not Linux ASR promotion. Use existing libraries within the stated support boundaries.

Prepare the source CLI

Python 3.10+ is required; CI uses 3.13. Clone recursively:

git clone --recurse-submodules https://github.com/NorthBoundWisdom/sonafolio.git
cd sonafolio
python3 configs/setup_local_app.py

Windows uses python instead of python3. Setup creates the default managed runtime/config without downloading weights.

source ~/Documents/SpeechToTextData/env/bin/activate
atot doctor
atot models

& "$env:USERPROFILE\Documents\SpeechToTextData\env\Scripts\atot.exe" doctor
& "$env:USERPROFILE\Documents\SpeechToTextData\env\Scripts\atot.exe" models

Activate Scripts/Activate.ps1 if shell policy permits, or keep using the full Windows path. A global Python/atot can be a different environment.

Build the current native app

With Xcode installed, from the repository:

python3 configs/ios_workflow.py build --platform mac --scheme Audio2TextApp \
  --configuration debug --distribution ad-hoc

Open build/app/SonaFolio.app. This is a local source build, not a new release. Rebuild/restart after updating source. See development for Qt/iOS.

Readiness checks

Check Expected result
Client Opens the chosen library
doctor Required backend, tools, weights ready
models Correct files under model/
Short Chinese task Transcript and QA without a cloud key

Models → · First transcript →