Buzz
Buzz 1.4.4 is a local Whisper-based speech transcription and translation tool for audio and video files and live microphone input, with TXT, SRT and VTT export.
What Buzz provides
Buzz can transcribe or translate audio and video files and can process microphone input. Local Whisper, Whisper.cpp or Faster Whisper models keep processing on the device after model download, while remote APIs or online-link workflows send content to the configured service.
Model, speed and storage trade-offs
Larger models may improve recognition but need more disk, memory and processing time. Start with a short sample, record the model and language settings and compare punctuation, speaker changes and terminology before processing a long recording.
Subtitle export and privacy
TXT, SRT and VTT are delivery files with different timing and formatting behavior. Keep source recordings, model cache and exports separate, and confirm the current backend before transcribing sensitive audio. Review date: 2026-08-23.
Save to your cloud drive
Open the cloud drive to get the file directly, or save it for convenient access on another device.
Quark Cloud Drive
RecommendedSave Buzz to this cloud drive
Baidu Netdisk
Save Buzz to this cloud drive
Buzz 1.4.4 first transcription and subtitle guide
Install the desktop build, choose a local model, transcribe a short sample and export SRT while observing how model size affects speed, memory and accuracy.
Before you start
- Prepare a short recording that you are allowed to process and that contains no sensitive information.
- Reserve space for the model cache and exports; larger models can require several gigabytes and more memory.
- Decide whether processing is local or uses a remote API before adding a recording.
Installation steps
- 01
Install the platform build
Use the Windows, macOS or Linux package, verify the version and review any operating-system signing prompt before launch.
- 02
Download or select a model
Choose a model matching language, speed and accuracy needs, wait for its download and note its local storage path.
- 03
Run a short transcription
Import a short file, select language and backend, start processing and compare the transcript with the source audio.
Quick start
- 01
Export subtitle formats
Export TXT for plain text, SRT for common subtitles or VTT for web video, then inspect timestamps and line breaks.
- 02
Compare model settings
Run the same sample with two models or language settings and record speed, memory, punctuation and terminology accuracy.
- 03
Archive the result
Keep the original, model name, settings, transcript and corrected subtitle together, with private recordings in a protected directory.
Usage tips
- Local model processing and remote API processing have different data paths.
- Model size affects disk, memory, speed and accuracy; test before batching.
- Subtitle timing still needs human review for names, punctuation and speaker changes.
Troubleshooting and uninstall
Why is transcription slow or out of memory?
Select a smaller model, shorten the sample, close other workloads and check available memory and disk. Re-run a short test before processing the full recording.
Why are subtitles out of sync?
Check the source frame rate, audio start offset, model segmentation and export format, then compare one short clip in the target player.
- Save transcripts and settingsExport corrected text and subtitles, record model and language settings and back up any project metadata before removal.
- Uninstall and clean model cacheRemove Buzz through the operating system and clear model or temporary files only after confirming that source recordings and exports are archived.
Frequently asked questions
Does Buzz require an internet connection for every transcription?
Local models can process on the device after download; model downloads, remote APIs and online-link features still need network access.
Does Buzz upload recordings automatically?
Local processing keeps audio on the device, while a selected remote API or plugin sends content to that configured service. Check the backend before sensitive work.
Does the download require an extraction code?
The Quark entry does not require one; the four-character code for the Baidu entry is shown beside its download entry.