Buzz Review

Buzz transcribes and translates audio offline on a personal computer, powered by OpenAI's Whisper. The README lists features: transcribing audio and video files or YouTube links, live real-time transcription from the microphone with a presentation window for events, speech separation before transcription to improve accuracy on noisy audio, speaker identification, multiple Whisper backends (CUDA acceleration for Nvidia GPUs, Apple Silicon support, and Vulkan acceleration for whisper.cpp on most GPUs including integrated ones), support for Hugging Face transformer model families, export to TXT, SRT, and VTT, an advanced transcript viewer with search and playback controls, keyboard shortcuts, a watch folder for automatic transcription of new files, a command-line interface, and a plugin system with plugins such as AI summaries. Recent versions require Apple silicon on macOS; the last version supporting Intel Macs is 1.4.5.

On macOS download the .dmg from SourceForge; on Windows download the installer from SourceForge, and note the app is unsigned so you must choose More info then Run anyway on the warning. On Linux use Flatpak, Snap, or an AppImage. You can also install from PyPI with pip install buzz-captions on Python 3.12, with ffmpeg installed, and run python -m buzz; Nvidia GPU support on Windows requires matching CUDA builds of torch. From version 1.4.6 onward the app can install CUDA acceleration from the About dialog. We did not install it.

A desktop GUI where you pick a file or a live input, choose a Whisper model and language, and run; results open in a transcript viewer with search, playback, and speed controls. The model you choose trades accuracy for speed and memory. This is based on the README only.

• Offline transcription and translation with Whisper
• File, YouTube link, and live microphone input
• Speaker identification and speech separation
• CUDA, Apple Silicon, and Vulkan acceleration
• Export to TXT, SRT, and VTT
• Watch folder, CLI, and plugin system

• Private by design because processing is local
• Multiple GPU backends make it usable on many machines
• Useful extras like watch folders and CLI
• Solid adoption at about 22,000 stars

• Unsigned Windows and macOS builds trigger security warnings
• Intel Macs are no longer supported by recent versions
• Accuracy and speed depend on local hardware and the model chosen
• Setup on PyPI with GPU support is fiddly

Buzz is free under the MIT license, with support through starring and donations. There is no paid tier.

Otter.ai is a hosted meeting transcription service with real-time collaboration, summaries, and integrations with calendars and video calls. Buzz is a local tool for transcribing files and live audio privately without a subscription, but it does not join your meetings or provide shared team workspaces.

• Journalists, researchers, and students transcribing recordings privately
• Creators who need subtitle files
• Anyone with a capable GPU or Apple Silicon Mac

• Teams needing meeting bots and shared notes
• Low-powered machines where large models are too slow
• Users who need a signed, enterprise-managed installer