Transkryptor

🇫🇷 Lire en français

A modern web platform for audio transcription, analysis, and synthesis — built on top of the Cloud Temple LLMaaS SecNumCloud-qualified API.

Table of contents

• Highlights
• Architecture
• Quick start
• Configuration
• Usage
• Project layout
• Roadmap
• License

Highlights

• Parallel audio transcription at scale. Files are chunked, transcribed in parallel via Whisper, and reassembled. Handles mono/stereo and common formats (MP3, WAV, M4A).
• Analysis & synthesis. The transcript is split into semantic batches, analyzed in parallel by a configurable LLM, then synthesized into a structured report.
• Five synthesis presets out of the box — executive summary, meeting minutes, action items, cleaned verbatim, thematic analysis — plus a fully custom prompt mode.
• Speaker diarization (LLM-based). Identifies turns of speech from the transcript with optional speaker-count hinting, streamed live as it runs.
• Multilingual interface: French and English UI; 15+ supported audio languages with auto-detection and an independent synthesis target language.
• lesur.ai visual identity: current v6.1 refresh with the cream/navy/cyan/amber design system, self-hosted Newsreader typography, and local runtime assets.
• Real-time observability: per-chunk progress grid, performance chart, live server logs over SSE.
• Cloud Temple LLMaaS backend: all calls go through the sovereign SecNumCloud-qualified Cloud Temple API. No third-party LLM providers in the data path.
• Secure API gateway: the API key lives server-side in environment variables and is never exposed to the browser.

Architecture

A Node.js/Express backend acts as an API gateway to the Cloud Temple LLMaaS endpoints (transcription, chat completions, diarization). The frontend is a vanilla-JavaScript single-page application served by the same Node process — no bundler, no framework.

browser ──► Node.js/Express gateway ──► Cloud Temple LLMaaS (Whisper + LLM)

Quick start

Prerequisites: Node.js 18 or higher, and a Cloud Temple LLMaaS API key.

Recommended macOS/Linux install:

curl -fsSL https://raw.githubusercontent.com/Lesur-ai/transkryptor/v6.2.0/scripts/install.sh | bash

The script installs the app into ~/Applications/Transkryptor on macOS and ${XDG_DATA_HOME:-~/.local/share}/transkryptor on Linux. It copies .env.example to .env when needed, runs npm ci, starts the service on http://localhost:3000, and opens the default browser.

Useful overrides:

curl -fsSL https://raw.githubusercontent.com/Lesur-ai/transkryptor/v6.2.0/scripts/install.sh -o /tmp/transkryptor-install.sh
TRANSKRYPTOR_INSTALL_DIR="$HOME/Transkryptor" \
TRANSKRYPTOR_REF="main" \
TRANSKRYPTOR_OPEN_BROWSER="0" \
bash /tmp/transkryptor-install.sh

On Windows, use WSL for now:

wsl bash -lc "curl -fsSL https://raw.githubusercontent.com/Lesur-ai/transkryptor/v6.2.0/scripts/install.sh | bash"

A native Windows installer should be a small PowerShell script that installs into %LOCALAPPDATA%\Transkryptor, runs npm ci, starts the Node service, and opens http://localhost:3000.

Manual setup:

git clone https://github.com/Lesur-ai/transkryptor.git
cd transkryptor
npm install
cp .env.example .env
# Edit .env — set CLOUD_TEMPLE_API_KEY and CLOUD_TEMPLE_ALLOWED_MODELS
npm start

Then open http://localhost:3000.

For development with auto-reload:

npm run dev

Docker

docker compose up --build

Configuration

All configuration lives in a .env file at the project root. Copy .env.example and fill in your values.

Variable	Required	Purpose
PORT	no (default 3000)	Port the server listens on.
CLOUD_TEMPLE_API_KEY	yes	Bearer token for the Cloud Temple LLMaaS API.
CLOUD_TEMPLE_ALLOWED_MODELS	yes	Comma-separated list of model IDs exposed in the UI. The order is also the display order, and the first entry is the default selected model.

Usage

1. Pick an analysis model from the sidebar dropdown.
2. Drop or pick an audio file (.mp3, .wav, .m4a).
3. Optional — pick an audio language (or leave auto-detect) and a target synthesis language.
4. Optional — enable participant detection if the audio has multiple speakers.
5. Click Start processing. Per-chunk progress, speed stats, and a performance chart light up as the work runs.
6. Once analysis is done, Synthesis becomes available. Pick a preset (or customize the prompt under Advanced prompts) and run it.
7. Use Download to save the active tab's content as Markdown.

Project layout

transkryptor/
├── src/
│   ├── client/
│   │   ├── css/
│   │   ├── i18n/             # FR/EN UI translations
│   │   ├── img/
│   │   ├── js/
│   │   │   ├── ui/           # UI modules (stats, chart, progress, results)
│   │   │   ├── analysisProcessor.js
│   │   │   ├── apiService.js
│   │   │   ├── audioProcessor.js
│   │   │   ├── main.js
│   │   │   └── …
│   │   └── index.html
│   └── server/
│       ├── logger.js
│       └── server.js
├── .env.example
├── docker-compose.yml
├── Dockerfile
├── scripts/
│   └── install.sh          # macOS/Linux one-line installer
├── package.json
└── README.md

Roadmap

• More audio/video container formats.
• Project-level user authentication.
• Persistent session storage for transcripts.

License

GPL 3.0 — see LICENSE.txt.

---

Built with ❤️ by lesur.ai.