Skip to content

Why repeated speech starts to sound like song

Loop a spoken phrase and it can start to sound sung, with no change to the audio. What makes a phrase turn, who hears it, and how to mine it for a hook.

· 8 min read Lesson 8 of 14 in Music Psychology

In short

  1. Loop a short spoken line exactly, with no pitch shifts or reordering, and hum the tune you start to hear: that tune already fits the words.
  2. Phrases with steady pitch inside each syllable turn into song most readily, so choose or edit spoken samples with that in mind.
  3. A melody you hear after an hour of looping may be inaudible to someone who hears the line once, so support it with an instrument or a sung version.
In this lesson (8)
  1. The voice memo that started singing
  2. A tool and a trap
  3. Exact repeats, level syllables and the listener
  4. Mine a spoken line for its tune
  5. Trusting a tune only you can hear
  6. Let the loop write the first draft
  7. Check yourself
  8. Sources

The voice memo that started singing

You pull a spoken line from an old voice memo for an intro and loop it while you build the beat. Twenty minutes later you cannot hear it as talking any more. The words seem to sit on notes, and you are humming them. Then a friend walks in, hears the loop twice and asks who is talking.

Hearing some spoken phrases over and over is enough to make them sound sung, with no change to the audio.

Why it matters

A tool and a trap

The effect gives you something useful. A looped spoken line can hand you a melody and a rhythm that already fit the words, which is hard to get by writing a tune first and forcing a lyric onto it. The lesson on lyric rhythm explains why that fit matters.

It also misleads you. After an hour of looping, you hear a tune that a first-time listener may not hear at all, so a vocal chop that sounds like a hook in your session can sound like stray dialogue to everyone else.

The science

Exact repeats, level syllables and the listener

Diana Deutsch came across the effect in 1995 while preparing the spoken commentary for a CD of musical illusions: a phrase from it, "sometimes behave so strangely", started to sound sung after she had played it several times. Deutsch, Henthorn and Lapidis (2011) then tested it. In the first experiment, listeners heard the phrase ten times and rated it on a five-point scale from "exactly like speech" to "exactly like singing". When every repeat was identical, the mean rating rose from about 1.3 to about 3.8, well past the midpoint. When the repeats in between were transposed by two-thirds of a semitone or more, up or down, the rating moved only slightly toward song, to about 1.9, and stayed on the speech side. When the syllables came in a jumbled order, the last hearing still sounded like speech. Neither condition produced the illusion, and the two did not differ reliably from each other.

Figure 1. Mean ratings in Deutsch, Henthorn and Lapidis's (2011) first experiment, read approximately from their Figure 2, on a scale from 1 (exactly like speech) to 5 (exactly like singing). The first and tenth hearings were the same recording in every condition. With exact repeats in between, the tenth hearing crossed into song. Slightly transposed repeats moved it a little toward song but left it on the speech side, and with jumbled syllables it barely moved.

In the second experiment, listeners heard the phrase once or ten times and then said it back. After one hearing they spoke it. After ten they sang it, and their pitches matched the original speech more closely than after one hearing. The sung versions were closer still to a simple tune in a key. Ten repeats left listeners with specific notes in their heads. Deutsch's listeners had musical training, but Vanden Bosch der Nederlanden, Hannon and Snyder (2015) replicated the effect in casual listeners with none, and confirmed that transposing the repeats disrupts it.

Not every phrase turns. Tierney, Dick, Deutsch and Sereno (2013) searched audiobooks and found 24 phrases that most listeners heard as song when repeated and 24 that stayed speech, both sets taken from the same three readers in the same proportions. The clearest acoustic difference was pitch movement inside syllables: the song-like phrases glided less. Their stressed syllables were only slightly more regular in timing, a difference that was not statistically reliable. Falk, Rathcke and Dalla Bella (2014) compared spoken phrases that differed in specific pitch and timing properties. Level pitch targets on the syllables made the transformation happen more often and sooner than making the steps between syllables match a musical scale, and recurring contrasts between long and short syllables helped. A regular beat within and across repeats did not help.

Figure 2. An exaggerated sketch of pitch over one five-syllable phrase. In Tierney and colleagues' (2013) audiobook phrases, the ones that turned into song moved on average 27.6 semitones per second within a syllable, against 40 for the ones that stayed speech. Level syllables give the ear notes to hold on to.

The listener matters as well. Margulis, Simchy-Gross and Black (2015) played 24 native English speakers, most without music training, a segment of under three seconds cut from the same passage read in seven languages, each repeated ten times. Languages that are hard for English speakers to pronounce shifted most toward song. English shifted least, and that change was not statistically reliable. Regular or irregular gaps between repeats made no difference. The ratings also show the size of the effect: on average, even the strongest shift stayed on the speech side of the scale. In a study of 20 adults, Jaisin and colleagues (2016) found a weaker effect in native speakers of tonal languages such as Thai and Mandarin, where pitch carries word meaning.

Figure 3. Mean ratings from Margulis, Simchy-Gross and Black (2015), on a scale from 1 (exactly like speech) to 5 (exactly like singing), before and after ten repeats. Languages that are hard for English speakers to pronounce shifted most. English shifted least, and even the largest shift stayed below the midpoint.

A reading that fits these results, and close to the one Margulis and colleagues give for the language effect, is that the more readily a phrase is processed as speech, the harder it is to hear as song. Level syllables then give the ear notes to hold. Each finding rests on small samples and a handful of phrases, so expect your own lines to vary.

Try it

Mine a spoken line for its tune

0 of 7 steps done

Common mistake

Trusting a tune only you can hear

After an hour of looping, a spoken sample sounds melodic to you, so you build the hook on it. Your listeners hear it far fewer times before the next section starts. If the tune matters, double it with an instrument, sing it, or let the sample repeat several times before the part that depends on it.

The second mistake is chopping a spoken phrase into a new order every bar, or pitching each repeat to follow the chords, and then wondering why it stopped sounding like a melody. Those edits are close to Deutsch's jumbled and transposed conditions. Keep at least one exact, untouched repeat if you want the effect.

Takeaway

Let the loop write the first draft

Before you write a melody for a line, speak it, loop it and listen for the tune it already has. Pick lines whose syllables hold their pitch, keep the repeats exact, and transcribe what you hum. Then check it on fresh ears: if the tune only exists after twenty loops, give it an instrument or a voice that carries it from the first bar.

3 questions

Check yourself

  1. You loop a spoken line as a hook and pitch each repeat a semitone up or down so it follows the chords. What do the studies predict for the melody you hoped people would hear in it?
  2. You have two spoken lines from one narrator to loop for a hook. In the pitch editor, line A slides up or down inside most syllables, while line B holds a fairly level pitch on each syllable and steps between them. Which is more likely to start sounding sung?
  3. You have looped a spoken sample for an hour and hear a clear melody in it. A friend hears it twice and says it is just someone talking. What is the most likely reason?

Sources

Show all 6
  • Deutsch, D., Henthorn, T., & Lapidis, R. (2011). Illusory transformation from speech to song. Journal of the Acoustical Society of America, 129(4), 2245-2252.
  • Falk, S., Rathcke, T., & Dalla Bella, S. (2014). When speech sounds like music. Journal of Experimental Psychology: Human Perception and Performance, 40(4), 1491-1506.
  • Jaisin, K., Suphanchaimat, R., Figueroa Candia, M. A., & Warren, J. D. (2016). The speech-to-song illusion is reduced in speakers of tonal (vs. non-tonal) languages. Frontiers in Psychology, 7, 662.
  • Margulis, E. H., Simchy-Gross, R., & Black, J. L. (2015). Pronunciation difficulty, temporal regularity, and the speech-to-song illusion. Frontiers in Psychology, 6, 48.
  • Tierney, A., Dick, F., Deutsch, D., & Sereno, M. (2013). Speech versus song: Multiple pitch-sensitive areas revealed by a naturally occurring musical illusion. Cerebral Cortex, 23(2), 249-254.
  • Vanden Bosch der Nederlanden, C. M., Hannon, E. E., & Snyder, J. S. (2015). Everyday musical experience is sufficient to perceive the speech-to-song illusion. Journal of Experimental Psychology: General, 144(2), e43-e49.

Next lesson · 9 of 14

Attention moves through a mix like a spotlightListeners follow one part of a mix at a time. How the ear sorts sound into streams, why a crowded chorus buries the vocal, and how to hand the spotlight on.

Previous: Why a song grows on you with repeated plays

Portrait of Virzy Guns

Written by Virzy Guns

Songwriter and producer, founder of Virzy Guns Production. Now building HealingWave.

Browse beatsAbout Virzy Guns

Later in Music Psychology

See all 14 lessons

  1. Why contrast makes the chorus feel bigA chorus only sounds big next to something smaller. Why constant density stops registering, and how to build relief into a verse without losing its energy.
  2. Music sets up tension before the lyric explains itLoudness, pitch and note density build tension before a word is sung. What tension research shows, and how to make a build work with the vocal muted.
  3. A familiar sound can carry a memoryTape hiss, an old drum machine or a familiar chord move can point a listener to another time. Why it works, why you cannot choose the memory, and how much to use.