ljudanteckning

Make your media library searchable: scan folders, split audio, transcribe in parallel across NVIDIA GPUs, then export subtitles and searchable text next to each original file.

Screenshots

Here’s what a real run looks like on a multi-GPU box: chunking with FFmpeg, parallel transcription workers, and live GPU telemetry ("transkriptor" was the work-in-progress name of the project).

More screenshots

Chunking stage

Finished outputs next to media

Goals

• Recursively discover media files (audio + video)
• Extract audio and chunk into short WAV segments (FFmpeg)
• Transcribe chunks in parallel across N GPUs (faster-whisper / CTranslate2)
• Merge timestamps correctly and export:
  • .srt
  • .vtt
  • .whisper.json
  • .txt
• Clean up temp files after success
• Nice terminal UX:
  • colored output + banner
  • interactive TUI mode (later)
  • batch mode (--nocli)

Requirements

Ljudanteckning is designed for NVIDIA GPUs and gets its best performance with CUDA. If you have multiple GPUs available, it can run several transcription workers in parallel.

Core requirements

• Linux
• Python 3.10+
• FFmpeg (ffmpeg and ffprobe)

GPU requirements (recommended / expected)

• One or more NVIDIA GPUs
• NVIDIA proprietary driver installed (so nvidia-smi works)
• CUDA runtime libraries compatible with your driver
  (installing the CUDA toolkit/runtime is often the simplest way to get these)

Optional

• NVML Python bindings for nicer GPU telemetry in the UI (pip install -e ".[nvml]")

Quick check

nvidia-smi

If that command works and your GPUs show up, you’re ready to run GPU-accelerated transcription.

Install (developer setup)

Arch Linux (fish)

# 1) System packages
sudo pacman -Syu
sudo pacman -S --needed git ffmpeg python python-pip nvidia-utils

# 2) GPU sanity check
# If this fails, fix your NVIDIA driver installation first.
nvidia-smi

# 3) Clone + virtualenv + editable install
git clone https://github.com/albinhenriksson/ljudanteckning.git
cd ljudanteckning

python -m venv .venv
source .venv/bin/activate.fish

python -m pip install -U pip
pip install -e ".[dev,nvml]"

ljudanteckning --help

Debian / Ubuntu (bash)

# 1) System packages
sudo apt-get update
sudo apt-get install -y git ffmpeg python3 python3-venv python3-pip

# 2) GPU sanity check
# If `nvidia-smi` is not found, install the NVIDIA driver/utils for your distro first.
nvidia-smi

# 3) Clone + virtualenv + editable install
git clone https://github.com/albinhenriksson/ljudanteckning.git
cd ljudanteckning

python3 -m venv .venv
source .venv/bin/activate

python -m pip install -U pip
pip install -e ".[dev,nvml]"

ljudanteckning --help

Fedora (bash)

# 1) System packages
sudo dnf install -y git ffmpeg python3 python3-pip

# 2) GPU sanity check
# If `nvidia-smi` is not found, install the NVIDIA driver/utils for your distro first.
nvidia-smi

# 3) Clone + virtualenv + editable install
git clone https://github.com/albinhenriksson/ljudanteckning.git
cd ljudanteckning

python3 -m venv .venv
source .venv/bin/activate

python -m pip install -U pip
pip install -e ".[dev,nvml]"

ljudanteckning --help

Quick start

Note: CLI flags shown below are implemented progressively while the pipeline is being built. Run ljudanteckning run --help to see what’s available in your installed version.

1) Create a local config (recommended)

cp ljudanteckning.example.ini ljudanteckning.ini

(Optional) sanity-check what Ljudanteckning will use:

ljudanteckning config

2) Transcribe a single file

ljudanteckning run --nocli "/mnt/media/Mr.Robot.S01E01.1080p.BluRay.mkv"

3) Transcribe an entire directory (recursive)

ljudanteckning run --nocli "/mnt/media"

4) Pick GPUs explicitly (multi-GPU)

Example with 4 GPUs:

ljudanteckning run --nocli --gpus "0,1,2,3" --jobs 4 "/mnt/media"

Example for an 11-GPU box:

ljudanteckning run --nocli --gpus "0,1,2,3,4,5,6,7,8,9,10" --jobs 11 "/mnt/media"

5) Override model / language

ljudanteckning run --nocli --model large-v3 --language sv "/mnt/media"

Outputs

For each media file, Ljudanteckning writes transcript/subtitle files next to the original, e.g.:

• Movie.mkv
• Movie.srt
• Movie.vtt
• Movie.txt
• Movie.whisper.json

Temporary chunks and per-chunk JSON are stored in a hidden work folder next to the file (by default: .ljudanteckning/) and are cleaned up according to output.cleanup in your INI.