VideoCaptioner + CrispASR (Integrated)

A desktop app for automatic video subtitling / captioning. This repository is the VideoCaptioner application (based on WEIFENG2333/VideoCaptioner, v1.4.2), integrated with CrispASR for local speech recognition / transcription.

Published so others can clone the project and run VideoCaptioner with CrispASR.

---

Repository layout

main.py              # Application entry point  ->  python main.py
requirements.txt     # Python dependencies
app/                 # VideoCaptioner source (PyQt5 GUI, core, threads, views)
resource/            # Assets, subtitle styles, translations, and bundled binaries
  └─ bin/CrispASR/   # Prebuilt CrispASR binaries (Windows: crispasr.exe, *.dll)
_build/CrispASR/     # CrispASR engine SOURCE (build it for non-Windows platforms)

What is NOT included (by design)

To keep the repository lean and avoid committing large binaries / private data, the following are intentionally not tracked (see .gitignore):

• runtime/ — the bundled portable Python runtime (recreate it yourself, see below)
• VideoCaptioner/, AppData/ — runtime data, caches, logs, downloaded models
• work-dir/ — local working media files
• CrispASR build/, build-cuda/, models/ and other build artifacts
• Installer / frozen-app files (*.exe installers, unins000.*, backup zips)

Because of this, AppData/ (settings, logs, models, cache) and work-dir/ are created at runtime. The app recreates the folders it needs on first launch.

---

Quick start (run from source)

1. Clone

git clone https://github.com/lwt127/VideoCaptionerCripASR.git
cd VideoCaptionerCripASR

2. Create a Python environment (Python 3.11 recommended)

Windows (PowerShell):

py -3.11 -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install --upgrade pip
pip install -r requirements.txt

macOS / Linux:

python3.11 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
pip install -r requirements.txt

3. Make sure ffmpeg is available

The app needs ffmpeg for audio/video processing. Install it system-wide (so it's on your PATH) or place the binaries in resource/bin/.

4. Run

python main.py

---

Transcription backends

VideoCaptioner can transcribe locally using several backends.

CrispASR (integrated, zero-setup ✨)

CrispASR works out of the box — no manual installation. In the 语音转录 (Transcription) tab choose CrispASR, pick a backend engine + model, then start the task. On first use:

• If the CrispASR engine binary is missing, it is automatically downloaded from the official GitHub Releases (CrispStrobe/CrispASR) into resource/bin/CrispASR/.
• GPU acceleration: if an NVIDIA GPU (CUDA) is detected, the app automatically downloads/uses the CUDA build of the engine for much faster transcription; otherwise it uses the CPU build.
• The selected model is automatically downloaded by CrispASR on first run (cached under C:\Users\<username>\.cache\crispasr\ on Windows).

So users only need to select a backend/model and press start — everything else (engine, GPU build, model) is fetched automatically.

Windows ships a self-contained prebuilt crispasr.exe (auto-downloaded if absent). To build manually for other platforms, see _build/CrispASR/ (build-windows.bat, CMake presets) and copy the binary into resource/bin/CrispASR/.

Available CrispASR backends (all auto-downloadable): Whisper, Parakeet (TDT/RNNT/CTC variants), FastConformer-CTC, Canary, Voxtral Mini 3B / 4B Realtime, Granite Speech, Qwen3-ASR, Mega-ASR, Fun-ASR Nano, SenseVoice, Paraformer-zh, Cohere, wav2vec2 / HuBERT / data2vec, omniASR (1600+ languages), FireRedASR2, GLM-ASR, Kyutai STT, Gemma4-E2B, MiMo-ASR, MOSS-Audio, VibeVoice, KugelAudio, Moonshine — plus selectable VAD methods (Silero / FireRedVAD / MarbleNet / Whisper-VAD).

Faster-Whisper (optional)

Invoked as an external program (Faster-Whisper-XXL), not a Python package. Download it and point the app to it via Settings, or place it under resource/bin/Faster-Whisper-XXL.

Models

CrispASR models download automatically (see above). WhisperCpp / Faster-Whisper models can be downloaded via the in-app model manager into AppData/models/.

---

(Optional) Recreate the portable runtime

The original distribution shipped a bundled runtime/ (portable Python). That is not committed here. To reproduce a fully portable build, recreate a Python 3.11 runtime and install requirements.txt into it, or just use the .venv setup above for development.

---

Components & licenses

• VideoCaptioner — see the upstream project: https://github.com/WEIFENG2333/VideoCaptioner
• CrispASR — see _build/CrispASR/LICENSE.

Note: This is an integration/distribution repository assembled from a local install. Some setup steps (runtime, models, optional backends) must be performed locally because those large artifacts are not committed here.