Buzz

Buzz transcribes and translates speech from recordings, media files and a microphone. It suits people preparing interview transcripts, captions, searchable notes or accessible live text. Local Whisper-family backends can process audio on the computer once the required models are available.

Transcription and captions

Import audio or video, select a model and language, run transcription or translation and review the result. Export plain text or timed SRT and VTT captions. The publisher documents a transcript viewer with search and playback controls, live microphone recording, speaker identification, a watched folder, a command-line interface and plugins.

Different backends and model families have different capabilities and hardware requirements. NVIDIA CUDA instructions are not Mac requirements. Large models can consume substantial memory and take longer; speaker labels and timestamps are estimates, not verified identities or guaranteed timing. Inspect generated captions and translated text before publishing them.

Installation and architecture

Copy Buzz.app from the official DMG into Applications, or use brew install --cask buzz. The checked 1.4.5 Apple silicon DMG is 469,325,527 bytes. This is the ARM64 artifact; the publisher identifies 1.4.5 as the last release with a separate Intel build. Do not mistake a newer Apple-silicon-only README policy for proof that 1.4.5 never had Intel support.

Homebrew declares macOS 12 or later for this distribution. Read-only inspection of the exact 1.4.5 DMG confirms an arm64-only main executable for Apple Silicon. Its Mach-O deployment minimum is 11.0, and Info.plist contains no LSMinimumSystemVersion field. The installation guidance retains the higher macOS 12 packaging requirement; the lower executable build value does not establish that every bundled dependency supports macOS 11. No extra formula dependency is declared. Python, pip and FFmpeg setup instructions are relevant to the alternative source/PyPI route, not an unconditional manual setup for the packaged Mac app. Additional models and media take disk space beyond the DMG size.

Network, permissions and account

Microphone permission is required for live input; imported recordings need access to the chosen files. Using a loopback audio device to capture other programs is a separate installation and configuration decision, not a built-in guarantee that Buzz records system sound. Optional folder watching should be restricted to a folder whose contents you intend to process.

A local-model workflow has no mandatory Buzz account and can operate offline after model downloads. Downloading models, importing an online URL, checking updates or using a remote provider or plugin needs networking. Optional API services require their own credentials and may transmit audio or text and incur charges. Local transcription does not imply every plugin is local or free.

License and data risks

Buzz is free and open source under MIT. Model weights, external media services and optional plugins have their own licenses and terms. Confirm rights to record and process the source audio. Recordings, temporary conversions, transcript exports and logs can expose private speech or identifying data. Sanitize bug reports and choose an appropriate retention policy for working files and backups.

Review plugins before enabling them: a summary or post-processing plugin can send transcripts to a remote service or change exported text. Preserve original recordings until you have checked accuracy, especially for names, numeric details and multiple speakers. Do not bypass macOS security protections merely because a downloaded application reports an installation warning; obtain the correct publisher package and follow verified support guidance.

Sources: Official repository, Installation documentation, Exact releases.