AI Speech Writer

Most speech tools stop at the page. You get a draft, you copy it somewhere, and the first time you hear it out loud is when you are standing in front of people — which is the worst possible moment to discover that your third sentence is unsayable in one breath.

This page joins the two halves. Step one writes a draft to your brief; you edit it until it sounds like you; step two turns it into audio you can play back on the walk to work. Rehearsing against a recording beats reading in your head, because reading in your head silently repairs every problem the delivery will have. Your eye skips the clause with no natural breath in it. Your ear does not.

It also catches the two failures that spoil otherwise good speeches: running long, and sentences built for the page rather than the mouth. A recording gives you a real duration instead of a guess, and it makes the unspeakable sentences obvious within about ten seconds of hearing them.

The single field that decides whether a speech sounds like yours is the details box. Everything else — occasion, tone, length — shapes the container. The details are the content.

Two or three concrete things beat a paragraph of adjectives. Not "he is a great friend" but "we met queuing for a five-a-side pitch and he was forty minutes late to his own stag do". Specifics are what an audience remembers, and they are the one thing a model cannot invent for you. Anything generic in the output is generic because the brief was.

A useful test before pressing the button: could this brief describe anyone else you know? If it could, it is not yet a brief — it is a category. One remembered Tuesday is worth more than five adjectives.

• How you know the person, in one line — this usually becomes the opening.
• One story with a place, a time or a number in it. Vague warmth is forgettable; a llama on the phone during the stag weekend is not.
• A running joke, a habit, a phrase they always say.
• Anything that must not be said. The draft is only as safe as the brief, and the writer cannot know what the room does not know.
• Who is in the room. A speech for colleagues and a speech for family are different speeches, even about the same person.
• What you want people to feel at the end. Not the message — the feeling. It decides the last line.

A page of A4 is not a unit of time, which is why speeches overrun. Length here is set in minutes and converted at roughly 130 words per minute — an unhurried pace with room for pauses, for laughter, and for the breath you will need.

That rate is deliberately conservative. People speak faster under adrenaline than in rehearsal, so a draft timed at a calm pace gives you slack rather than a deficit. If you know you speed up when nervous, treat the number under your draft as a ceiling; if you tend to slow down, aim short and let the room stretch it.

The counter updates as you edit, so you can watch a three-minute speech become a five-minute one while you are adding to it — which is exactly how speeches grow past their slot without anyone noticing.

Almost nobody is asked to speak for longer than they can hold a room, and almost everybody writes past it anyway. As rough targets: a toast under a minute; a best man or maid of honour speech three to five; a eulogy three to five; a retirement or farewell tribute around five; an award acceptance under two; a wedding speech from a parent three to four.

Work speeches are governed by the slot rather than the form. A business update fills the time it is given and stops; a keynote is written to its slot minus two minutes, because two minutes always disappear into the room. If you have been given twenty, write eighteen.

When in doubt, take a minute off. No audience has ever complained that a speech ended too early, and the shorter version is nearly always the better-written one — length is usually where the padding hides.

Treat what comes back as a strong first draft, not a finished speech. The edit is where it becomes yours, which is why the draft box is a text editor rather than a read-only result.

Read it aloud once before changing anything, and mark every place you stumble. A stumble is almost never your fault — it is a sentence written for the eye. Cut it in two and the problem usually disappears.

Then do a pass for words you would not use. Everyone has a register, and a draft that reaches above yours is what an audience hears as inauthentic long before they could say why. Swap the words that are not yours for the ones that are, even where the replacement is plainer.

• Replace any sentence you stumbled over with two shorter ones.
• Cut the throat-clearing at the start. Speeches almost always begin a sentence or two later than they were written to.
• Keep the specifics, cut the abstractions. If a line could sit in anyone else's speech, it is doing no work in yours.
• Say the last line out loud several times. It is the part people carry out of the room, and the part most often left unrehearsed.
• Put a paragraph break where you want to pause. The generator reads those breaks as breath, so the audio rehearses your pacing too.

It will not leave you blanks to fill in. Placeholder brackets are how people end up reading "[NAME]" out loud at a wedding, so the writer is instructed to work around a missing detail rather than mark it. For the same reason it will not invent a name for you: a speaker who introduces themselves under the wrong name is worse off than one who does not introduce themselves at all.

It will not write in the voice of a real public figure, and it will not help present generated words as someone else's own. A speech is a claim about who is speaking, and that is exactly the thing worth keeping honest.

And it will not know the room. It does not know which uncle is not speaking to which cousin, what the company announced last week, or which joke stopped being funny in March. That knowledge is yours, it never reaches a draft unless you put it in the brief, and it is usually the difference between a speech that works and one that merely reads well.