Rule of thumb: around 40 words a minute typed, around 150 spoken
That ratio is workshop knowledge, not a study. Your own typing speed decides how much of it applies to you.
Time saved
Skald turns speech into finished text inside whichever app you are already in. Hold a key, speak, let go.
Windows on ARM64 (Snapdragon) →
Two honest numbers, and one thing we will not pretend to know.
That ratio is workshop knowledge, not a study. Your own typing speed decides how much of it applies to you.
In 2016 researchers from Stanford, Baidu and the University of Washington put speech recognition up against a keyboard. Dictating English was three times faster than typing, with about 20 % fewer errors. They measured on smartphones rather than at a desk, so take it as the order of magnitude, not as your personal result.
Every minutes-saved calculator on the internet runs on numbers somebody invented. What you actually save depends on how much you write in a day and how much of it you would have reworked afterwards. Try it on your own texts.
In the small texts you write all day and never count.
You already know what it should say. Speak it, and Skald sets the greeting, the paragraphs and the closing line in the register the thread is written in.
Speak the first version. What arrives is a message and not a transcript, and it reads like chat instead of like a memo.
Talk through what was decided while it is still fresh. Skald turns the run-on speech into paragraphs and bullet points you can hand on.
In an editor Skald keeps its hands off your identifiers and your quotes. It writes what you said, punctuation included, and changes nothing else.
Plain dictation hands you back what you said: the false starts, the “no, wait”, the sentence you began twice. Cleaning that up by hand costs roughly what typing it would have cost.
Skald reads the app, the window and the field before the text is written. A message to a colleague comes out short. A first mail to a stranger comes out with a greeting and a closing line. Code stays exactly as you spoke it.
Microsoft Word ships its own Dictate button, no extra software required. How to turn it on, and where it falls short →
Text appears at the cursor. Cleaned in tone for the app you're in — Slack reads casual, Mail reads polished, code stays untouched.
Select text, hold the key, say what to change. "Shorter." "More formal." "As bullets." "In English." The selection is replaced — no diff to review.
When you have more to say than fits in a field. Opens a Skald window, captures everything, cleans it into structured prose you can drop where you want.
Put them in your dictionary once, straight from the menu bar. Product names, client names, the jargon of your field, spelled the way your team spells it. The dictionary stays on your machine.
German in, German out. English in, English out. If you switch mid-sentence, the sentence stays switched.
Wherever a cursor blinks: mail, chat, your editor, the address bar, a terminal, a web form. The rare app that blocks accessibility access is the exception.
Manifesto
Three sentences about privacy soothe no one. Four architectural decisions do.
Skald never writes audio files to disk. Not as a cache, not as a debug artefact. What you say is heard once — and then it's gone.
Audio leaves your machine only to be turned into text. Our own servers are in the EU. For speech recognition and AI cleanup we work with specialised partners — each one named in the privacy policy, each one with model training on your data turned off.
Your voice is never used to train anyone's models. We pay the surcharge to keep your speech out of training data — that's on us, not on you.
No analytics pixel in the product. No crash reporter. No "anonymous" heartbeat from your Mac, Windows, or Linux app. Your audio stays in memory and is never written to disk.
You get 7 days of the full Pro feature set. Skald shows the trial-day counter right in the app and emails you a reminder three days before it ends. If you don't actively switch to Pro, your account drops to the free plan — no hidden auto-charge.
macOS 13+ and Windows 11 (x64 and ARM64) are available to download today. Both clients ship the same feature set — Dictate, Command, Brain-Dump, the personal dictionary, settings. macOS is the oldest channel and the default route; Windows is now stable. The Linux client is built natively on GTK4 and is in closed beta; join the waitlist for early access.
Audio is only in memory during a recording — never written to disk on your machine or on our servers. Transcripts pass through cleanup and are not logged by default. Our own servers are in the EU. For speech recognition and cleanup we use specialised AI partners with model-training opt-out; they are all named in the privacy policy.
Yes. Go to heyskald.com/billing → Manage subscription and cancel in one click. Cancellation takes effect at the end of the current billing period; Skald keeps working until then. No lock-in, no retention hotline.
Free plan, no card. The account is created from inside the app once you have installed it.