On Windows, Buzz is distributed as a Windows X64 zip of about 2.7 GB. Extract the archive and run the setup program; because the app is unsigned, Windows SmartScreen may warn you, and the installation notes specify choosing “More info” and then “Run anyway.” Plan for the large download and enough disk space before starting.
Buzz is an open-source desktop application under the MIT license, with versions for Windows, macOS (Apple Silicon) and Linux. During first use, it downloads the Whisper model you select. Smaller models process faster, while larger models can provide more accurate results; large models may run slowly without a GPU. Once a model is available, transcription can take place on your computer without an internet connection.
After setup, Buzz can transcribe or translate audio and video files, as well as audio from YouTube links. It supports Whisper, whisper.cpp with Vulkan acceleration on most GPUs including integrated ones, CUDA on NVIDIA cards, and Whisper-type models from Hugging Face. The selected backend and model determine how the local transcription work is performed.
From recording to transcript
For recorded material, Buzz can separate speech from background noise before transcription and identify speakers in the resulting recording. Speaker labels help distinguish voices in a multi-person conversation, while the transcript viewer provides search and playback controls for reviewing specific passages against the audio.
Transcripts can be exported as TXT, SRT or VTT. TXT is suited to plain text records, while SRT and VTT preserve subtitle timing for video workflows. These formats let users move from speech recognition to text review or subtitle preparation without relying on a separate transcription step.
Live captions and repeatable work
Buzz can also caption microphone input in real time. A presentation window displays captions for events or other situations where spoken content needs to be shown as it is delivered. This live mode differs from processing a saved file, but it uses the same general purpose of turning speech into readable text.
For recurring tasks, a watch folder can handle files placed in a designated location, and the command-line interface supports scripted operation. Plugins extend the workflow, including options such as AI summaries. These tools are useful when transcription is part of a repeated process rather than a one-off manual session.
Local processing and practical fit
Buzz runs its Whisper-based transcription and translation locally, so audio is not uploaded as part of that work. An internet connection is needed to download models, and processing audio from YouTube links also requires internet access. This distinction matters when handling private recordings or working offline after model setup.
The combination of file transcription, live microphone captions, speaker identification, subtitle export and automation makes Buzz relevant to journalists, students and video makers who need private transcripts or subtitles. Its main setup considerations are the substantial Windows package, the SmartScreen warning and the processing demands of larger models, particularly on systems without a GPU.