Whisper runs inside the app
HookWise runs Whisper through whisper.cpp on your computer, using Metal on a Mac. The audio is read straight from your file and turned into text, and every word gets a start and end time. The transcript is saved in the project, so captions, filler removal and AI Clipping all work from the same text.
You choose from three models. Base, about 148 MB, ships with the installer. Small, about 488 MB, and Medium, about 1.5 GB, download inside the app from Hugging Face, and HookWise checks the download before using it. You can delete a downloaded model later to free disk space.
Choosing a model for your recording
Start with Base on a clean recording, such as a podcast mic or a lapel mic in a quiet room. Move up to Small or Medium when the room echoes, music plays under the voice, the speakers have strong accents, or the topic is full of technical terms. Larger models are usually more accurate and take longer on the same computer.
Set the language yourself when you know it. Auto-detect is handy for unknown files, and a fixed language can prevent a wrong guess when a video opens with music or a greeting in another language. For a YouTube video that already has auto-captions, import those and you can skip transcription.
What to expect on speed and accuracy
Transcription works offline on every plan, and nothing is uploaded. You only need the internet to download Small or Medium, or to import from a link. How long a transcript takes depends on the video length, the model and your computer. HookWise publishes no speed figures, so test a typical episode during the 7-day trial.
No model gets every word right. Check names, brand terms and lines where languages mix, then edit them in the transcript panel. As long as the number of words stays the same, each word keeps its timestamp, so your captions stay in sync after the fix.
