Great Scott Labs

Vocabulary

Say your biggest customer's name three times.
Count the spellings.

That one test tells you more about a dictation tool than any review will. Run it on whatever you use now, before you read the rest of this page.


Why the other tools can't fix this

Every dictation tool offers a dictionary. Almost all of them work the same way: you tell it that when it writes X, it should write Y instead. A find-and-replace on the output.

That only works if the tool gets it wrong the same way every time. And with a name it has never heard, it doesn't. Here is a real result — the word "Fastenal", said three times into a $29 local competitor:

Fasten all Fasten all that's in all

There's nothing stable to map. By the third attempt it isn't even close. A correction table needs a predictable mistake, and unknown proper nouns never fail the same way twice.

What Lickety does instead

It moves the fix upstream. Before the recognizer listens to a word, Lickety hands it the list of names you've told it about — spelled the way you spell them. The recognizer isn't guessing at sounds and hoping; it already knows those words exist and picks them.

You never have to know what it got wrong. You just type the word once.

To be recorded

Screen capture — 20–25 seconds

Say a name, get it wrong. Type it into the Vocabulary list. Say it again, get it right. One continuous take, no cuts, no voiceover.

Screen capture slot
click-to-play, poster frame required
not autoplay — it has a beginning

A brand animation would be wrong here. The claim is "you can do this in thirty seconds," so the evidence has to be the actual thirty seconds.


The measured result

On a 64-sentence test set of real customer and vendor names, seeding the vocabulary took the error rate on names from 11.1% to 4.8%, and to 3.2% with the phrasing tuned. On a separate sentence containing sixteen real business names, the best run was 16 out of 16.

One voice, one test set, one machine — your numbers will be different. The mechanism is the part that transfers.

Names

Customers, vendors, coworkers. The words you retype every day because nothing has ever spelled them right.

Places

Small towns and regional names that no training set has enough of. Natchitoches and Schuylkill come out as something else entirely until you say otherwise.

Internal systems

The acronym only your company uses. It has no public presence anywhere, which is exactly why no model knows it.