iPhone and Android · in development
Speak without notes.
Notes are a crutch for missing structure. ColdSpeak trains the thing underneath them — so you can stand up with nothing in your hands and still know exactly where you are going.
No lessons. No praise. A coach that tells you exactly where you were vague, and one thing to fix before the next rep.
The coach
It will not tell you that you did well.
Every verdict is about the speech, never about you, and every one ends with a single thing to do differently. There is no praise sandwich and no score padded to spare your feelings. If you want encouragement from a practice tool, this is the wrong one. If you want to know exactly where you lost the room, it is the right one.
It counts your hedges
“I think”, “sort of”, “maybe” — every one found in the transcript and shown to you, because each takes the weight out of the claim it sits in.
It says when it found no point
If nothing you said can be restated as one claim, the coach says exactly that, rather than inventing something generous.
One thing to fix, then it stops
Six scores, measured delivery, and a single instruction for the next rep. Never a list, never a pep talk.
Your first score will be low. Everyone’s is. The second rep is where it starts to move, and that is the number worth watching.
The premise
Nobody great was remembering sentences. They were composing live.
A speaker who works without notes is not running a better memory. They are holding a small number of durable structures, and a deep supply of concrete material they can reach for. The speech is assembled in the room.
Which means the skill is trainable, and it is not the one most people practise. Rehearsing a talk until it is smooth teaches you that talk. It does not help you the moment someone asks a question you did not plan for.
What it trains
Four muscles, not one.
Most speaking apps work on the fourth and leave the other three to confidence. The first three are what let you throw the notes away.
- 01
Structure
Eight skeletons, each a handful of beats. Small enough to hold under pressure, general enough to carry any topic.
- 02
Substance
Your own bank of stories, figures and analogies, drilled on a recall schedule. This is what actually replaces notes.
- 03
Pressure
Two hostile questions after every rep, answered with no preparation. This is where speakers without notes get exposed.
- 04
Delivery
Pace, pauses, fillers, sentence length — measured from the recording, not guessed at.
The loop
Ten minutes. Five steps.
- 01
Draw a prompt
Something you have not rehearsed. Uncomfortable on purpose — practising only on your own material is rehearsal, not training.
- 02
Thirty seconds
Pick a skeleton, pick three specifics. No writing. If you cannot hold it in your head now, you will not hold it on your feet.
- 03
Speak
A ring divided into your skeleton’s beats shows whether you are still on the beat you think you are on.
- 04
Get marked
Six rubric scores, measured delivery, and exactly one instruction for the next rep. Never a list.
- 05
Take the questions
Two hostile follow-ups, specific to what you actually said. Answer them cold.
The hardest question it asks
“Nothing in this rep could be restated as a single claim.”
After every rep, ColdSpeak tries to state your point back to you in one sentence. When it cannot, it says so rather than inventing something generous. That verdict is uncomfortable and it is the most useful thing the app produces: it means you spoke about the topic without arguing anything — the failure that hides best behind a confident delivery.
What it measures
Numbers, not impressions.
Delivery is computed from the timing of every word you speak. The same speech always produces the same numbers, so when the line moves you know you moved it.
- Pace
- 130–160 wpm
- Words per minute, and how much that swings.
- In range, a listener follows without effort. Faster reads as nerves and gives them nothing to hold; slower reads as searching. Big swings are the tell of composing on the spot.
- Fillers
- under 2 / min
- “Um”, “uh”, “you know” — counted per minute of speaking.
- Each one is a sound made instead of a pause. Under two a minute nobody notices; past that it is all they notice. The fix is the pause itself: silence reads as thought.
- Pause ratio
- 12–28%
- The share of speaking time that was silence.
- In range is a speaker who stops to let a point land. Under it, nothing lands; over it, you are hunting for the next thing and the room can hear it.
- Dead air
- none over 3s
- The single longest silence.
- Up to three seconds is weight — the room leans in. Past three, it decides you have lost the thread. Silence on purpose, never by accident.
- Sentence length
- 12–22 words
- Average words per sentence.
- In range can be held in the ear. Longer, and the listener forgets how the sentence started; shorter throughout, and you sound clipped and cannot build.
- Hedges
- 0–1
- “I think”, “sort of”, “maybe”, “kind of” — counted.
- Each one takes the weight out of the claim it sits in. One can be honesty; several, and nothing said committed to anything.
- Specifics
- 4+ / min
- Named, checkable things per minute — figures, dates, people, places.
- Four a minute is the difference between an opinion and a case. Only real ones count: a made-up figure is worse than none.
Two registers
Pick what gets rewarded.
The same rubric, weighted differently. Train the voice the room needs.
Direct
Blunt, numerate, consequence-first
Short words. Hard figures. Plain causal claims. Hedging is marked down hard, and an abstraction with no number attached counts as a failure. Rewards sentences a listener could disagree with, because they commit to something.
Reflective
Analogical, historical, calm
Framing that earns its place, analogy that illuminates rather than decorates, and movement between levels of abstraction a listener can follow. Deliberate pauses read as command here, not as hesitation.
Early access
Ten minutes a day, and no notes.
ColdSpeak is in development for iPhone and Android. Leave an address and we will write once, when it is ready to use.
One email, no list sharing, unsubscribe in a click.
Questions
+Is the coach harsh?
Direct, not harsh. It marks the speech, never the speaker, and it never pads a low score with encouragement. People who want to be told they did well find it cold. People who want to know exactly where they were vague find it the most useful ten minutes of their day.
+Is this a public speaking course?
No. There are no lessons to watch. It is a training loop you run for about ten minutes a day, the way you would train anything else physical.
+Do I need an audience?
No. You need a phone and somewhere you are willing to speak out loud. The hostile questions supply the pressure that an audience normally would.
+What happens to my recordings?
Speech is transcribed so it can be measured and marked. Read the privacy page for what is kept and for how long — it is written plainly and it is short.
+Will it make me sound like someone else?
It should not. The registers bias what gets rewarded, not what you say. The fundamentals — one clear claim, real evidence, a stated consequence — are the same in any voice.