Whisper on your desktop, not someone else's server

What it does Buzz is a desktop application for transcribing and translating audio using OpenAI’s Whisper model. It runs entirely offline on macOS, Windows, and Linux, handling files, YouTube links, or live microphone input. The project also exposes a CLI and supports watch-folder automation for batch processing.

The interesting bit The value isn’t the model—it’s the packaging. Buzz adds speaker identification, speech separation for noisy audio, a presentation-friendly live-transcription window, and GPU acceleration across Nvidia CUDA, Apple Silicon, and even Vulkan for integrated GPUs. The “Advanced Transcription Viewer” with search and playback controls turns raw Whisper output into something you can actually work with.

Key highlights

• Multiple Whisper backends with broad GPU support, including Vulkan for non-gaming GPUs
• Speaker diarization and pre-transcription speech separation
• Export to TXT, SRT, and VTT formats
• Watch folder for automatic processing of new files
• Unsigned Windows binaries (the README warns you’ll need to click past SmartScreen)

Caveats

• Windows installer is unsigned, which may give enterprise users pause
• GPU support on Windows requires manual CUDA dependency management for PyPI installs
• The README doesn’t clarify whether speaker identification and speech separation use bundled models or require additional downloads

Verdict Worth a look if you need Whisper’s accuracy without shipping audio to a cloud API—journalists, accessibility workers, or anyone with sensitive recordings. If you already have a comfy Python environment and don’t need a GUI, the raw Whisper repo is probably enough.