The Dream Is Not the Peak

By Noam Shemla ·

By Noam Shemla · Speeches That Lasted

TL;DR We measured the speech Martin Luther King Jr. gave at the March on Washington on 28 August 1963 with the same instrument our members record into. This time we had real audio, decoded from the recording itself, alongside the transcript. Pace still bottoms out at our band’s floor of 50 (102 words a minute, elapsed time still counting every second of applause). What’s new is the acoustic data: his pitch does not peak during the eight sentences that open “I have a dream.” It climbs after that passage, through “let freedom ring,” and peaks in the closing “free at last.”

Every entry in this series so far has published either a score that was wrong (Selma), a score with nothing to compare it to (Gettysburg, and Churchill’s missing litany), or a structural finding that matched the history cleanly (the first version of this entry, built from text alone). This version has real audio behind it for the first time, and the acoustic data complicates the clean story.

A speech still under copyright

This speech is still very much in copyright. The King estate has been unusually active in enforcing it (famously winning a federal case over unauthorised commercial use of the recording), and it remains protected in the US until 2059. So, as with Churchill above, this entry works from structure, timing and short attributed fragments instead of the text in full.

It’s also the first entry in this series with no video figure, despite film of the speech existing. The house rule for this series is to embed only from the rights holder’s own upload, never a reupload of uncertain standing. The Selma article documents the same diligence for its own video. We were not able to confirm a YouTube upload that clears that bar for this speech, so none is embedded here. The measurements below do not depend on it.

Short, even sentences until the end

The shape shows a speech that keeps its sentences short and even for most of its length, then breaks that pattern exactly where the repetition figure below shows the anaphoric runs landing: a scatter of very short bars right where “I have a dream” and “let freedom ring” open sentence after sentence in quick succession.

What the repetition figure caught

That clustering deserves a closer look. A phrase repeating three or four times scattered across a long speech (which is what Selma’s repeated openings mostly look like) reads as a returning motif. Eight repeats of the same words opening eight sentences in close succession, with almost nothing interrupting, is a different device: a run, and our figure shows it as one, a dense block rather than scattered marks. “Let freedom ring” does the same thing immediately afterward. The speech doesn’t sprinkle its anaphora. It detonates it, twice, back to back, in its final third.

A pace score that counts the crowd

This is Selma’s blind spot again, in miniature. A quarter of a million people stood on the National Mall responding to this speech as it was delivered, and a pace score computed from elapsed time has no way to separate King’s voice from the crowd’s. The number is a true statement about the recording and a misleading one about the speaker.

Where the pauses and the pitch part ways with the legend

Thirty-two real pauses over 0.5 seconds, timed from the gaps between spoken words instead of inferred from punctuation. The single longest, 13.8 seconds, falls around 7 minutes in, well before the “I have a dream” run starts at roughly the 10-minute mark; the pace timeline shows the same stretch as one of the slowest 60-second windows in the whole speech (18 words a minute at t=270s), most likely a sustained ovation the pause detector reads correctly as silence on King’s part. The fastest 60-second window in the entire address, 204 words a minute, doesn’t fall during “I have a dream” either. It’s roughly 14 minutes in, inside the “let freedom ring” litany.

  • 274 Hz mean pitch. ±70 Hz, clearly not monotone by our threshold.
  • 0.09 avg volume. Range 0 to 0.42 across the full recording.
  • 32 real pauses. 27 read as deliberate (1.5s+), 1 as hesitant.
  • 13.8s longest pause. Before the “I have a dream” run begins.

The acoustic axes this entry didn’t have before, measured from the recording.

One finding complicates the clean story the text-only version of this entry told. Windowing the raw pitch readings into 30-second blocks across the whole speech, the mean sits between 240 and 290 Hz for most of the address, including through the eight sentences of “I have a dream” itself, which read no higher than the surrounding material. The pitch climbs afterward: the last four 30-second windows of the recording, covering “let freedom ring” through the closing “free at last, thank God almighty, we are free at last,” run 288, 297, 309 and, in the final window measured, 321 Hz, the highest of the entire address. The acoustic peak of this speech, by our instrument’s own numbers, is not its most quoted line. It’s the one that comes after it.

We take that at face value instead of forcing it to match the legend. “I have a dream” is the rhetorical center of gravity. It is the device every textbook points to, and our own repetition figure finds it instantly. But rhetorical structure and vocal delivery are different signals, and here they peak at different moments. If anything, the data suggests King built to his acoustic climax gradually, across the whole closing third of the speech, with “I have a dream” serving as the structural turn and the vocal summit arriving later.

Limits the audio did not remove

Three more recordings, and the widest range in the corpus

We later measured three more King recordings, all of them the same kind of occasion as this one: a live address to an audience. They produced the largest spread of any speaker we have measured, and the two extremes sit twelve months apart.

The tempting explanation is length, since longer recordings score lower throughout this corpus. It does not survive contact with these two: both are long, and the higher-scoring one is barely shorter. What separates them is register. One is read from a prepared text on a policy question; the other is preached, largely without notes, in the cadence he grew up inside. The instrument does not know a word of the history. It hears the difference anyway.

Two lessons for your own recordings

There are two lessons here. The pace score is still the least trustworthy number in this article. It answers a question (how fast did the room move) that a member’s own dashboard is usually not asking (how fast did YOU talk). And the new acoustic data is a caution against assuming a speech’s rhetorical high point and its vocal high point are the same moment. Here they weren’t. If your own Practice Studio report flags a repeated phrase as your speech’s structural turn, that’s a useful signal about your content. Whether it’s also where your voice is doing the most work is a separate question, and one to check instead of assume.

The same audio pipeline our members record into: word-timed pace and pauses from timed captions, pitch and volume from the decoded recording.

Measured

  • Word count and pace against real elapsed time
  • Real pauses (0.5s+) from gaps between timed words
  • Pitch: mean 274 Hz, ±70 Hz, not monotone
  • Volume: average 0.09 across a 0 to 0.42 range
  • Filler and hedging counts and rates
  • Vocabulary diversity: 40.7 percent of words distinct
  • Positions of every sentence-opening phrase that repeats three times or more

Inferred, not measured

  • Word timings are spread evenly inside each caption cue, so pauses inside a single cue can’t be seen
  • Elapsed time includes the audience’s applause and response, which our pace clock cannot separate from the speaker’s voice
  • Sentence boundaries follow the punctuation of the timed captions

Sources