VoiceSnap ProvsOpenAI Whisper

Whisper takes audio and returns text. That is the whole interface. Everything else you associate with dictation is somebody's code wrapped around it.

Consider what has to exist between “Whisper is very good” and “I dictated this email”. Something has to hold a global hotkey across every application. Something has to open the microphone, buffer audio, and decide when you stopped speaking. Something has to run the model without freezing your laptop. Something has to decide that the four seconds of silence in the middle was a paragraph break and not the end. Something has to strip the filler words. And then something has to insert the text into the focused field of whichever application has it — which behaves differently in a browser, a terminal and a native text view.

That list is the product. It is also the part that takes months and never quite ends, because every application handles focus and text insertion slightly differently.

Building it yourself is a genuinely reasonable choice if you are the sort of person who enjoys it. Plenty of engineers have a hotkey script that pipes audio through a local Whisper build, and they are happy with it.

The costs are the ones you would expect and a few you would not. Model files take gigabytes. Startup latency has to be hidden or you feel it every time. Transcription competes with your build for CPU. It breaks after an OS upgrade. It works in your editor but not in that one Electron app. And the whole thing exists on one machine, so setting up a second laptop means doing it again.

None of that is hard exactly. It is just work that never finishes, and it is the reason paid dictation apps exist at all when the underlying model is free.

If you have a pile of recordings — interviews, lectures, podcast episodes, support calls — and you need transcripts, Whisper is the correct tool and VoiceSnap Pro is useless to you. VoiceSnap Pro does not accept audio files at all; it only listens live.

If the audio must never leave your machine, Whisper running locally is the strongest guarantee there is, stronger than any policy a cloud service can offer.

And if you need to transcribe at volume on a server, or fine-tune on domain audio, or translate speech to English, those are things an open model does and a closed dictation app does not.

No offline mode, no local model, no file transcription, no batch processing, no fine-tuning, no source code you can read. It is a closed desktop app for macOS and Windows and it has not shipped yet.

What you get in exchange is the last mile, already built and maintained: a shortcut that works everywhere, cleanup tuned for writing rather than for a transcript, a vocabulary list you can edit, a searchable library, and someone to email when it breaks on your machine.