A great coach doesn't just hear your words — they hear the music underneath. Here's what they're actually listening for.
I Once Watched a Coach Change a Speech in Twelve Words
Years ago, I sat in on a rehearsal where a student was practicing Nehru's "Tryst with Destiny." He had the words perfect. Every line memorized. Pacing steady. No fillers. By any checklist, it was a solid delivery.
The coach listened with her eyes closed. When he finished, she said: "You delivered every word at the same altitude. The speech has mountains in it. Give me the mountains."
That was it. No notes about WPM. No filler count. No completion percentage.
She'd heard something no transcript could capture — and it changed everything about what he did next.
The Gap Between Correct and Compelling
Most speech feedback lives in the world of correctness. Did you hit the target pace? Did you say "um" too often? Did you cover the full script?
These things matter. They're the foundation. But they're also where most feedback stops — because they're easy to measure.
A great speech coach hears past correctness into something harder to name. They hear the music underneath the words. The rise and fall of pitch that signals conviction. The moment your voice thins out because you're reading instead of feeling. The pause that lands like a thunderclap versus the pause that sounds like you lost your place.
This is the difference between a speech that is delivered and a speech that is felt. And for a long time, I assumed it was impossible to teach a machine to hear it.
I was wrong — but getting there required understanding exactly what a human coach is actually listening for.
The Five Things a Great Coach Hears
After spending months talking to speech coaches, debate instructors, and voice teachers, I started to notice a pattern. Their feedback — no matter how differently they phrased it — kept circling the same five dimensions.
1. The Energy Arc
A coach hears where your energy goes across the full length of a speech. Are you building toward something? Peaking in the middle and fading? Flat from start to finish?
The best speeches have a shape. Churchill built relentlessly toward his climaxes. Obama's cadence would dip into quiet intimacy before surging. A coach hears this arc instinctively — and when it's missing, the speech feels like it's going nowhere.
Technically, this maps to amplitude over time. We divide a speech into thirds — opening, middle, closing — and measure the average energy in each. A "building" arc means your closing energy exceeds your opening by at least 30%. A "fading" arc means the opposite. "Flat" means you stayed at one level throughout.
Most speakers fade. They start strong and run out of gas. A coach catches this in real time. Now, so can software.
2. The Pause Taxonomy
Not all silences are created equal. A coach hears three completely different things in a pause:
Natural pauses — 1.5 to 3 seconds. Breathing. Regrouping. Normal and expected.
Deliberate pauses — 3 to 6 seconds. These are the weapon of great speakers. Martin Luther King held silence before "I have a dream" long enough to make an audience of 250,000 lean forward. A deliberate pause signals control, not absence.
Hesitation pauses — 6 seconds or longer. This is where a coach's eyebrow goes up. The speaker has lost their place, their thought, or their nerve. These silences don't feel dramatic. They feel uncomfortable.
A great coach doesn't just count your pauses — they classify them. And they'll tell you that a speech with zero pauses is worse than a speech with deliberate ones.
3. Intonation — The Melody of Conviction
This is the big one. The thing that separates a technically correct delivery from one that moves people.
Intonation is the rise and fall of your pitch as you speak. It's the melody of your voice. And it carries enormous emotional information.
When someone truly believes what they're saying, their pitch moves. It rises on points of emphasis. It drops into gravity on moments of weight. It inflects — changes direction — at transitions between ideas.
A monotone speaker can have perfect words, perfect pace, perfect adherence to the script. But without pitch variation, the speech lands flat. A coach hears this instantly. They'll say things like "you're reading, not speaking" or "I need to hear that you care about this line."
What they're actually hearing is a narrow pitch range — your voice staying within 3-4 semitones when an expressive speaker would use 10 or more. They're hearing an inflection rate that's too low — your pitch moving in one direction for too long instead of dancing around key words.
We measure this with a technique called autocorrelation — detecting the fundamental frequency of your voice 75 to 400 times per second. From that signal, we extract pitch range (in semitones, because human ears perceive pitch logarithmically), variation, contour direction, and inflection rate. The result is an intonation score that correlates remarkably well with what a human coach would call "expressive" versus "flat."
4. Projection — Are You Filling the Room?
A coach can tell in seconds whether a speaker is projecting or merely talking. Projection isn't volume — it's the quality of the sound. A projected voice has brightness and presence. A under-projected voice sounds muffled, thin, or swallowed, even if the microphone picks it up fine.
This maps to something called spectral centroid — essentially, where the center of gravity of your voice's frequency spectrum sits. A well-projected voice pushes energy into higher harmonics, giving it cut and clarity. A mumbled voice concentrates energy in the low-mid frequencies.
We also measure spectral flatness, which tells us how tonal versus noisy a voice sounds. Clear, articulate speech has distinct tonal peaks. Muffled or breathy speech flattens out into noise.
Together, these give us a projection score and a clarity score — and they catch exactly what a coach means when they say "I need more from you" or "open up your voice."
5. Dynamic Range — The Distance Between Loud and Quiet
A coach listens for contrast. The speakers who hold attention aren't the loudest — they're the ones who move between loud and quiet with purpose.
Think of it like music. A song played at one volume is boring, no matter how good the melody is. The same is true of speech. A 2x dynamic range — meaning your loudest moments are twice the energy of your quietest — signals an expressive, engaging delivery. Below 1.5x, and the coach hears monotony.
Why This Matters More Than Word-Perfect Delivery
Here's what I've come to believe after building this: two speakers can deliver the same speech with identical words, identical pace, identical filler count — and one will move you while the other puts you to sleep.
The difference is entirely in these five dimensions. The energy arc, the pause quality, the intonation, the projection, the dynamic range. The music of the speech.
Traditional speech metrics — WPM, completion, adherence — tell you if someone did the assignment. Delivery metrics tell you if someone performed it.
This is why a score of 100% used to bother me. Two students, same speech, both "technically perfect." But play back the recordings and the gap is undeniable. One recited. The other inhabited the words.
Teaching a Machine to Listen Like a Coach
The challenge wasn't the signal processing — though getting reliable pitch detection from a browser microphone has its own adventures. The real challenge was weighting these signals the way a human coach would.
A coach doesn't score you on five axes and average them. They have an intuitive hierarchy: intonation matters more than amplitude, deliberate pauses matter more than total pause count, projection matters most when combined with pitch variety.
Our vocal variety score reflects this. When all analysis tiers are available, intonation carries 35% of the weight. Projection and spectral variety carry 25%. Dynamic range and amplitude variation carry 30%. Pause mastery carries 10%. These weights came from dozens of conversations with coaches about what they prioritize — not from an algorithm optimizing for a metric.
And because not every device can run full spectral analysis, the system degrades gracefully. Full analysis on capable hardware. Pitch-plus-amplitude on mid-range devices. Amplitude-only as a baseline that always works. The coach gets quieter on less capable hardware, but never goes silent.
The Feedback That Actually Changes Behavior
The most important thing I learned from watching real coaches: they don't give you a number. They give you a picture.
"Your energy faded in the last third." "You held one pitch for the entire second paragraph." "That pause before the closing line — that was powerful. Do more of that."
This is why our AI coach doesn't just report scores. It receives all of these signals — the energy arc shape, the pause classification breakdown, the intonation score, the projection level, the pitch contour — and translates them into the kind of paragraph a great coach would speak to you after a run-through.
Not "your vocal variety score is 62." But "your opening had genuine energy, but you flattened out during the second half — the closing deserves the same fire you brought to the opening. And that 4-second pause before the final line? Keep it. It's doing exactly what you want it to do."
That's the feedback that makes someone want to try again. Not a dashboard. A voice that heard you — really heard you — and told you what to do next.
Your voice has music in it. The question is whether you're playing it — or just reading the notes. Practice a speech in The Thinking Speaker, and find out what a coach actually hears.