Transcribe & Truchement

Your language toolbox on a USB stick — no internet, no account, no data leaving your computer.

Two portable desktop applications that work entirely offline:

• 🎙️ Transcribe — turn audio and video into text and subtitles
• 📖 Truchement — translate text and documents, and look up words in monolingual dictionaries

Both run from a regular folder or directly from a USB drive. Everything they need — the Python environment, the speech and translation models, the dictionaries — lives next to the executable. Plug the stick into another machine and everything still works, even without a network connection.

---

Why offline?

• Privacy — your recordings, documents and searches never leave your machine. Nothing is uploaded, ever.
• Availability — works on a train, in the field, behind a strict firewall, or anywhere without connectivity.
• No lock-in — no account, no API key, no subscription. Free and open source (MIT).

Screenshots

Transcribe	Truchement

---

What can I do with it?

🎙️ Transcribe

• Subtitle a video in minutes — drop in a video, get a clean .srt file, and optionally embed English + translated subtitle tracks into a single .mkv.
• Dictate a text file — record yourself and get a .txt transcript.
• Transcribe into another language — Whisper can transcribe directly to English, or you can translate the subtitles afterwards.
• Clean up difficult audio — built-in loudness normalization, noise reduction and volume adjustment before transcription.
• Grab online media for later — download a video or audio source (via yt-dlp) while you do have a connection, process it offline afterwards.

Typical pipeline:

[Video file]
     │
     ├─ 🔍 Analyze ──────────────── audio info + levels
     ├─ 🎚️ Preprocess ───────────── loudnorm / denoise / volume
     ▼
 📝 Transcribe (Whisper) ───────── movie.en.srt
     ▼
 🌐 Translate subtitles ────────── movie.fr.srt   (or any installed language)
     ▼
 📦 Mux with FFmpeg ────────────── movie.mkv  (both subtitle tracks embedded)

📖 Truchement

• Translate text between dozens of language pairs (English, French, Spanish, German, Portuguese, Italian, Russian, Japanese, Chinese, Arabic and more), powered by Argos Translate — the same open models used by LibreTranslate.
• Translate whole documents — .txt, .srt, .docx, .pdf, and .html (HTML keeps its structure, styles and scripts intact; only visible text is translated).
• Look up definitions in offline monolingual dictionaries (English, French, Spanish, German, Italian, Portuguese… and more available), with full-text search. Select a word in a translation and jump straight to its definition.
• Interface in your language — the UI adapts to your system locale and can be switched at any time.

---

Quick start

The easy way — with Baluchon

Baluchon is a small companion launcher that downloads the heavy parts for you (Whisper models, dictionaries, translation packages, the Python environment) and can create desktop shortcuts — handy when the apps live on a USB stick.

1. Download the latest release of Transcribe/Truchement from the Releases page and unzip it where you want it to live (folder or USB drive).
2. Run Baluchon in that folder and let it fetch the components listed in downloads.csv.
3. Launch transcribe or truchement. Done.

The manual way

1. Unzip a release (or clone this repository) into its final location.

2. Create the embedded Python environment (this installs Argos Translate and the language packages into the app folder — it will not touch your system Python):

   cd build/python
   ./setup_venv_lang.sh

3. Download one or more Whisper models into build/whisper/models/ — pick according to your machine:

   Model	Size	Good for
   ggml-base.bin	~150 MB	quick drafts, modest hardware
   ggml-medium.bin	~1.5 GB	good accuracy / speed balance
   ggml-large-v3-turbo-q8_0.bin	~0.9 GB	best quality on capable machines

4. Download the dictionaries you want into build/dic/ (one file per language, ~SQLite databases):

   en-en · fr-fr · es-es · de-de · it-it · pt-pt

   More languages (ru, ro, da, el, ja, tr, sv, zh…) are listed in downloads.csv.

5. Launch the app:

   ./truchement     # translation & dictionaries
   ./transcribe     # transcription & subtitles

Windows: support is in active development. The apps are designed to run from the same folder layout on both systems; a Windows build will be published in the releases.

---

Running from a USB stick

This is a first-class use case, not an afterthought:

• Copy the whole application folder to the stick — models, dictionaries and the Python environment travel with it.
• Nothing is written to the host machine's user profile; language packages and settings stay inside the app folder (build/argos-data/, build/dic/…).
• On Linux, prefer copying with cp -a or rsync -a. On file systems that don't support symbolic links (exFAT, FAT32, NTFS), make sure the embedded environment was created in "copies" mode — the release archives are prepared this way.
• If something was lost in a copy, Truchement includes a built-in diagnostic that tells you exactly what is missing and how to fix it, instead of failing silently.

How it stays self-contained

transcribe/                    ← the whole app: copy this folder, and it works
├── transcribe, truchement     ← the executables
├── build/
│   ├── python/venv/           ← embedded Python env (never touches system Python)
│   ├── whisper/models/        ← speech-to-text models (*.bin)
│   ├── argos-data/packages/   ← installed translation packages
│   ├── dic/                   ← offline dictionaries (*.db)
│   ├── ffmpeg/                ← media toolbox
│   └── yt-dlp/                ← media downloader
├── i18n/                      ← interface translations
└── *.sh                       ← command-line shortcuts for power users

Dictionaries are plain SQLite databases — a table entries {lemma, definition} plus an entries_fts full-text index — so you can build your own or convert existing ones (e.g. with PyGlossary).

Command-line shortcuts

Everything the GUIs do is also scriptable. A few examples:

Script	What it does
audio2srt.sh	audio/video → subtitle file
audio2en.sh	audio in any language → English text
traduire-srt.sh	translate an existing .srt
soustitre.sh	full video → subtitled video pipeline
creerMKV.sh / inclure-srt.sh	embed subtitle tracks into an .mkv
enregistrer.sh	record from the microphone
download_url.sh	fetch online media for offline processing

Troubleshooting

• "No languages installed" in Truchement — the translation packages are missing from build/argos-data/packages/. Re-run build/python/setup_venv_lang.sh (or Baluchon), or use the in-app Diagnose button to see where the app is looking.
• First translation is slow — the first use of a language pair loads its model into memory (a few seconds). Subsequent translations are near-instant.
• App works on one machine but not after copying to a stick — the copy probably dropped symbolic links. Recopy with cp -a / rsync -a, or use the release archives, which are link-free.

Roadmap

• ✅ Linux support
• 🚧 Windows build (single stick usable on both OSes)
• 🚧 In-app management of translation packages and dictionaries
• 💡 Automatic source-language detection
• 💡 More dictionaries

Suggestions and bug reports are welcome in the issue tracker.

Built on the shoulders of

• whisper.cpp — offline speech recognition
• Argos Translate — offline neural translation
• FFmpeg — audio/video processing
• yt-dlp — media downloading
• PyGlossary — dictionary conversion

License

MIT — see LICENSE.

---

« Truchement » is an old French word for an interpreter — someone who carries meaning between languages. That's exactly what this toolbox does, and it fits in your pocket.