Jonathan Secora Pearl, "Eavesdropping with a Master: Leoš Janáček and the Music of Speech", Empirical Musicology Review 2006:
The composer Leoš Janáček (1854-1928) has been noted for his interest in speech melodies. Little discussion has focused however on the field methods that he used in gathering them, nor on the products themselves. Janáček spent more than three decades, transcribing thousands of what he termed nápěvky mluvy [tunelets of speech] in standard musical notation. The record that remains of these efforts is impressive both for its volume and its quality, as well as for its potential to reveal aspects of the perceptual overlap between music and language.
Here's an edited and typeset example of one of his "tunelets": According to Pearl, this one
is dated 18 November 1897, presenting us the scene of an old woman at the butcher’s. Já ty paznehty nemám ráda [I don’t like those hooves] she says. Janáček provides the following background: Žena stará – nechtěla asi koupit si “nožičky” u řezniku – mluví k druhé ženě – zvolna – klidně [An old woman – it seems she didn’t want to purchase “little feet” at the butcher’s – she speaks to a second woman – slowly – calmly].
Pearl provides this audio file of (a piano version of) the tune:
Apparently Janáček heard the intonation of Czech phrases as a sequence of musical notes, or could persuade himself that he did so. I've never shared that impression — my experience is much more like that of Joshua Steele, who wrote (Prosodia Rationalis: Or, An Essay towards Establishing the Melody and Measure of Speech, to be Expressed and Perpetuated by Peculiar Symbols, 1775):
[T]he melody of speech moves rapidly up or down by slides, wherein no graduated distinction of tones or semitones can be measured by the ear; nor does the voice (in our language) ever dwell distinctly, for any perceptible space of time, on any certain level or uniform tone, except the last tone on which the speaker ends or makes a pause. For proof of which definition we refer to experiment, as hereafter directed.
Whilst almost every one perceives and admits singing to be performed by the ascent and descent of the voice through a variety of notes, as palpably and formally different from each other as the steps of a ladder; it seems, at first sight, somewhat extraordinary, that even men of science should not perceive the slides of the voice, upwards and downwards, in common speech. […]
We have accents in English, and syllabic accents too ; but there is no change of the tone in them ; the voice is only raised more, so as to be louder upon one syllable than another. — That there is no other difference is a matter of fact, that must be determined by musicians. Now I appeal to them, whether they can perceive any difference of tone betwixt the accented and unaccented syllable of any word? And if there be none, then is the music of our language, in this respect, nothing better than the music of a drum, in which we perceive no difference except that of louder or softer. [as quoted by Steele — the original seems to be slightly different.]
It's certainly true that most people are surprisingly bad at characterizing the "tunes" of speech as a time-function of pitch, even in such simple matters as whether a syllable is rising or falling, though Burnett seems to have been unusually imperceptive.
Steele trained himself to transcribe such pitch contours precisely by removing the frets from his viol, imitating a remembered speech-melody by slides of his finger on the string, and then notating the movements in terms of a modified form of traditional musical notation, in which
instead of using round or square heads for the notes to be marked on this scale (as in the ordinary music) let us substitute sloping or curving lines, such as expression may require; […] which lines, when drawn one the foregoing scale, will easily shew through how many quarter tones the voice is to slide; and these I call the accents or notes of melody.
Here's an example of what the results look like: Thanks to Google Books, you can now read Steele's 1775 work from beginning to end, an experience that I heartily recommend. And while there are probably benefits to doing your intonational analysis on the viol, it's easier these days to use free software like Praat or Wavesurfer.
So why did Janáček perceive a "graduated distinction of tones or semitones" where Steele (and I, and modern pitch-tracking software) hear only "slides"? I don't think that Czech was sung back in 1900, "through a variety of notes, as palpably and formally different from each other as the steps of a ladder" — it's certainly now no less slide-y than English is.
Perhaps Janáček had a form of synesthesia within a single sense. Or maybe he just forced himself to answer the question, "If I had to notate this as a song, what would be the best way to do it?"
Some LL posts on related topics: