Install Buzz on Mac, Windows or Linux, pick the right Whisper model, and fix the common failures: stuck on loading model, blank transcript, GPU not used, crashes on long files.
Mac: brew install --cask buzz or the DMG from GitHub releases. Windows: the .exe installer from GitHub (Microsoft Store build lags). Linux: pip install buzz-captions or the Snap/Flatpak.
Whisper small is the sweet spot for most laptops — tiny/base are fast but miss words; large-v3 is accurate but needs ~10 GB RAM and is slow on CPU.
File → Import, choose the audio, set the language (or auto), enable word-level timestamps if you need SRT/VTT export.
Same Whisper-class accuracy, no install, no model downloads, no GPU questions. Speaker labels, word timestamps and every export (TXT, SRT, VTT, DOCX) included. Free instant preview; credits from $2 and they never expire.
First run downloads the model (75 MB–3 GB) with no progress bar on some builds — give it time on a slow connection. If it never finishes, delete the model cache (Mac: ~/Library/Caches/Buzz, Windows: %LOCALAPPDATA%\Buzz) and retry.
Usually the wrong audio device or a silent/DRM-protected file. Play the file in a normal player first; if it plays, switch the Whisper model from tiny to small — tiny collapses on noisy audio.
The default build is CPU-only on Windows/Linux. GPU needs the CUDA build of whisper.cpp under the hood — on Apple Silicon, Metal is used automatically on recent versions; update if you're below 1.0.
Known pattern on 8 GB machines with medium/large models — memory. Use small, or split the file, or run the file through a hosted engine instead.
Real-time needs tiny or base; anything larger falls behind on CPU. For meeting-length live captioning a streaming service holds up better.
File → Export after selecting the transcription. TXT, SRT and VTT are supported; word-level SRT needs word timestamps enabled BEFORE transcribing.
More on Buzz: the full Buzz page · Buzz alternatives · all transcription tools