# Same sentence, only the melody changes

> In a question, the voice rises at the end. And the end of the sentence is where generated voices fail most.

Prosody · October 28, 2026 · Pedro Cormann, Founder & Machine Learning Engineer · https://lero.studio/research/en/melodia

In Portuguese, a yes or no question can use the same words as a statement. "Tá tudo certo." (All good.) and "Tá tudo certo?" (All good?) differ only in melody, the pitch contour across the sentence. In the readings below, the statement rises in the middle and falls at the end; the question stays almost flat and jumps up on the last syllable.

_(Interactive on the page: https://lero.studio/research/en/melodia)_

## The end of the sentence

The end of a sentence carries much of its meaning: a statement falls and fades, a question rises, and human voices often finish in creak (vocal fry). We compared the pitch of each generated sentence with a human reading of the same sentence, aligned in time, looking at the last second on its own. The generated voice has a narrower melody (7.4 vs 10.0 semitones), creaks less at the end (21% vs 33%) and, when it goes wrong, tends to end higher, as if the sentence were not over. Listen to the pairs.

_(Interactive on the page: https://lero.studio/research/en/melodia)_