ClipCast.News Transcriber 1.4.0

Windows: open ClipCast.News-Transcriber-1.4.0-Windows-x64.exe.
macOS: extract the ZIP, move ClipCast.News Transcriber.app into Applications, and open it. Choose Apple Silicon (arm64) or Intel (x64) for your Mac.
Linux: extract the tar.gz and launch clipcast-news-transcriber. A graphical desktop with Tk/X11 libraries is needed; tested target is Ubuntu 24.04 x64. From a terminal: ./clipcast-news-transcriber

Acceleration: compatible NVIDIA GPUs on Windows/Linux x64 are selected automatically when at least 1.5 GB GPU memory is available. First GPU use downloads about 1.1 GB of checksum-verified NVIDIA libraries into the private app data folder; no system driver is changed. GPU failures fall back to CPU processing. Other systems use 1-4 CPU threads depending on available processors. The small speech model, timestamps and evidence checks are preserved. The status panel shows which processor is in use.

Appearance: use the Light mode / Dark mode button at the top right, or Alt+T. Your choice is saved on this computer. The layout adapts to smaller windows; scroll to view transcript evidence.

First launch downloads a checksum-verified portable Python runtime and the free speech engine. Setup progress opens in your browser; the transcription interface is a desktop window. No CMD launcher or manual Python installation is required. Allow 9 GB disk space and 8 GB RAM. Close memory-heavy apps before processing. Speech recognition uses thirty-second chunks and a smaller model with automatic low-memory fallback. Speech and news-writing models run in separate processes, one at a time. Models download on first use (about 3 GB total).

Before transcribing, click Connect for news review. Sign in on the website, compare the displayed device code, and explicitly allow automatic transcript uploads. Choose a supported creator and a video/audio file or supported public video link. Audio/video processing takes place on your computer. The app automatically sends the AI-rewritten article, original transcript text, title, original video URL (or a generic local-recording label), and creator selection to the site's system administrators. It never uploads the recording itself. Local TXT, JSON and SRT exports remain in Documents/ClipCast.News Transcripts.

Each transcript creates a private draft story. Nothing publishes until a system administrator reviews and approves it. Local AI rewrites the full transcript into a news article before submission. If writing fails, the original transcript stays saved and the app retries; no unfinished draft is submitted. Creator labels do not prove speaker identity. Verify important quotes, context, names, negation and claims against original recordings.

If an upload fails, the transcript remains in the local outbox and retries while the app is running. Use Retry pending uploads if needed. A connection lasts 90 days and can be revoked at the website's /connect page. No new transcription starts without an active connection. Once started, a job can finish offline; its transcript waits to upload. Rate limit: 100 submissions per account per day; larger backlogs retry later.

Restricted videos, DRM, platform bot checks and some video links may be unavailable; choose an authorized local file instead. Some YouTube links require a separately installed Deno runtime. Recordings over 24 hours must be split. No software guarantees accurate transcription of every recording.

These builds are unsigned. The Mac app is not Apple-notarized and has not been run on a Mac here; macOS may block it until the user explicitly permits this trusted download in Privacy & Security. Windows may show an unknown-publisher warning. Do not disable system security protections. Windows x64 and Linux x64 are tested; Mac and Linux ARM builds are provided as unverified builds.

Runtime setup files and setup.log: Windows %%APPDATA%%/ClipCast.News Transcriber Native; macOS ~/Library/Application Support/ClipCast.News Transcriber Native; Linux ~/.config/ClipCast.News Transcriber Native. Models/outbox use the ClipCast.News Transcriber folder under local app data (Windows) or ~/.local/share (Mac/Linux). Upgrades reuse existing CreatorLens data folders to preserve models, connections, and pending uploads. Earlier exports remain in Documents/CreatorLens Transcripts; new exports use Documents/ClipCast.News Transcripts. No service or login task is installed. Desktop tokens are Windows-user-encrypted on Windows and stored with owner-only permissions on Mac/Linux.

NVIDIA runtime license files are included in the local nvidia-runtime-v1 folder.

Third-party licenses are included with downloaded runtime/dependencies:
Python: https://www.python.org/psf/license/
Portable Python: https://github.com/astral-sh/python-build-standalone
faster-whisper (MIT): https://github.com/SYSTRAN/faster-whisper
CTranslate2 (MIT): https://github.com/OpenNMT/CTranslate2
PyAV (BSD): https://github.com/PyAV-Org/PyAV
yt-dlp: https://github.com/yt-dlp/yt-dlp#license
Speech model: https://huggingface.co/Systran/faster-whisper-small
News AI (Apache 2.0): https://huggingface.co/Qwen/Qwen3-4B-GGUF
llama.cpp (MIT): https://github.com/ggml-org/llama.cpp
