Nell's Echoes
Log in
Back to the journal

Teaching the tool my low register: Étude a, before and after

The tool that paints my voice was almost completely unfamiliar with my lower register: it had learnt my vowels in the mid and high registers. I sang scales and sustained notes in the lower register so that it could learn them; the ‘Étude a’, repainted, shows what has changed.

Étude a : avant et après que l'outil apprenne mon grave

What I did

The tool recognises my vowels based on what it has learnt from my recordings. Below C4, it had almost none: barely 1 per cent of its training data. I recorded my seven vowels in the lower register, in scales from C♯3 to E4 and as sustained notes, and it learnt them in addition to what it already knew. I then had ‘Étude a’—my first composition written according to a set of grammatical and prosodic rules—repainted using the same recordings and the same transduction rule: the only thing that changes is what the tool hears in my lower register.

What I observed

On sustained notes in the lower register, the tool now recognises the sung vowel 87 times out of 100 (previously 35); on recordings it had never heard before, it loses nothing (from 89 to 90 per cent).

In ‘Étude a’, below C4, the written vowel is recognised three times out of four (previously once in four). In voice 3, the lowest, recognition rises from 30 to 77 per cent, and that of ‘ou’ from 41 to 88 per cent. Above E5, recognition remains low (29 per cent): the limit is no longer due to the tool, but to the physics of the sound.

What I make of this

In the high register, a soprano tunes the first resonance of her mouth to the note: the voice gains in power, but the vowels sound similar; above E5, my ‘i’, ‘e’ and ‘u’ become ‘é’. In the lower register, on the other hand, my vowels are very much present: it is the tool that did not recognise them. The tool cannot paint what it does not recognise; but it can learn, and that is what this experiment demonstrates. It also means that the tool evolves alongside my voice.

Questions

What remains for it to learn from my voice?

To what extent does a vowel sung in the high register remain the same vowel?

References

Elodie Joliveau, John Smith, Joe Wolfe, ‘Tuning of vocal tract resonance by sopranos’, Nature, 2004. Johan Sundberg, The Science of the Singing Voice, 1987.