AudioTranscriptor

AudioTranscriptor is a desktop app for people who record Italian conversations and need them written down: journalists, researchers, lawyers, podcasters, students. Drop an audio or video file, get an editable Italian transcript, translate it to English when you need to, and export TXT, SRT or DOCX.

Whisper runs on your machine. There is no account, no upload of your media, no telemetry, and the app works offline apart from the two things listed in Privacy.

Download

Release preparation status: the macOS Apple silicon package has been built and locally tried. The Windows installer and portable zip are built and self-tested by the GitHub Actions workflow (packaging, legal files, self-test that runs a real transcription); the Windows interface itself has not yet been run by hand on a Windows machine. Use https://aversa.design/audio-transcriptor/ for official download information; no release has been published yet, so do not assume the files below are downloadable.

File	What it is	Use it when
AudioTranscriptorSetup-1.0.2.exe	Windows installer, per-user, no administrator rights	Recommended on Windows. Start menu entry, uninstaller, license shown during setup
AudioTranscriptor-Portable.zip	The same Windows app as a folder	You cannot install software on the PC, or you want to run it from a USB stick
AudioTranscriptor-1.0.2-arm64.dmg	macOS app bundle (Apple silicon)	Recommended on a Mac. Drag it to Applications
SHA256SUMS.txt	Checksums of every file	You want to verify the download

The Windows builds are identical inside; the installer only adds shortcuts and the uninstall entry. Windows builds are unsigned; macOS builds are signed ad-hoc, not Developer ID-signed or notarized, so either system may display a warning. Only proceed after verifying the source and checksum. See INSTALLATION.md for first-launch instructions; security approval steps vary by OS version.

Screenshot

Features

• Local transcription with faster-whisper (CTranslate2 + Silero VAD): MP3, WAV, M4A, MP4, MOV, MKV.
• Model choice with honest sizes: small (about 0.5 GB, default), medium (~1.5 GB), large-v3 (~3 GB). Downloaded once, cached, reusable offline.
• Cancellable downloads: Cancel Process answers while the model is downloading, not after.
• Listen while you correct: a player with the spoken sentence highlighted, "Play from cursor" (Ctrl/Cmd+J), Esc to play/pause, Ctrl/Cmd+[ and ] to move 5 seconds.
• Editable transcript: fix names and numbers before translating; what you see is what gets exported.
• Subtitle export that respects your edits: lines are redistributed over the cues by timing, and a cue whose sentence you deleted is emptied instead of resurrecting the old text.
• English translation through the OpenAI API, in batches of 20 segments, with a per-segment fallback when a batch answer cannot be matched one-to-one, so translations never land on the wrong subtitle.
• Confirmation before anything leaves the computer: before translating, the app shows how many sentences and characters will be sent and to which model. It can be turned off in Settings.
• Measured progress: minutes of audio processed and sentences translated, no invented percentages.
• Six exports: Italian/English × TXT, SRT, DOCX.
• Interface in English, Italian, German or French: follows the system language on first launch, switchable in Settings without restarting. The audio language (Italian) and the translation target (English) do not change with it.
• Light and dark themes, background processing, drag and drop.

Privacy

The short version, and the file with the details is PRIVACY.md:

Data	Where it goes
Audio and video	Stays on your computer, always
Transcript and edits	Stay local unless you choose translation
Transcript text	Sent to api.openai.com only when you press Translate to English, with your own key
Whisper model	Downloaded once from Hugging Face (model name and IP only, never your media)
Settings and API key	Local settings file, owner-only permissions
Usage data, crashes, analytics	Not collected

No account, no sign-up, no server of mine in the middle. If you cannot upload transcript content, use the app without the translation feature: transcription and exports work fully offline.

Three honest caveats: the OpenAI key is stored in plain text in the settings file (see PRIVACY.md for the location and how to delete it), automatic transcription can be wrong, and the app comes with no guarantee of security and no obligation of support or updates — TERMS.md explains the responsibility that stays with you and the limits of that disclaimer.

How it works

media file ──► faster-whisper (local, CTranslate2 + Silero VAD)
                    │
                    ├─► editable Italian transcript ──► TXT / SRT / DOCX
                    │
                    └─► [optional] OpenAI API, batches of 20 segments ──► English SRT / TXT / DOCX

Alignment between the edited text and the subtitle cues lives in audio_transcriptor/alignment.py; the exporters never guess. A cue that lost its sentence is exported empty, not with stale text.

Getting started

python -m venv .venv
source .venv/bin/activate          # Windows: .venv\Scripts\Activate.ps1
pip install -r requirements.txt
python -m audio_transcriptor

Python 3.9 or newer runs from source, but 3.9 is end of life (October 2025), so prefer 3.11 or newer. The Windows build runs on 3.12; build_macos.sh prefers 3.13, 3.12 or 3.11 and only falls back to the system python3 with a warning.

Run the tests (no model download, no API key needed):

python -m unittest discover -s tests -t . -v

This is the beginning of the README. Read the rest at the original. Read the full README ↗