In terms of the correct pronunciation for perdoni, I would say that the stress should definitely be on the second syllable (perdòni).
As you suggested (well, I interpreted it as a suggestion ), I’ve listened to the voices for the multitude of sentences containing perdoni. Some of them are Giorgio, and some are Carla, some are stressing on the first syllable, some are stressing on the second syllable.
However, for those that are stressing on the first syllable (irrespective of whether it is Giorgio of Carla), I don’t hear perdoni, but rather perdone. Could you try listening again and see what you think?
– Warning!! - What follows is “intelligent” speculation … –
Moreover, I’m going to suggest that in the case of perdoni there may be a good reason why this causes the algorithms some problems in knowing how to put the correct stress on the word, and that is because there is a similar problem with perdono.
So in the case of perdono, there are in fact two perfectly good pronunciations with different stresses, one on the first syllable, and one on the second syllable -
-
perdòno - verb: first person singular present indicative of perdonare - or alternatively - noun: the translation of “forgiveness”
-
pèrdono - verb: third person plural present indicative of perdere
However, if you search for sentences that contain perdono in Clozemaster, the stress seems almost random across the sentences, with some getting the correct pronunciation, but many sentences being in error. This isn’t restricted to Clozemaster either. Google translate won’t put the stress on the first syllable if I type in “loro perdono la partita” and hit the Text to Speech button, instead it puts the stress on the second syllable . If I ask Reverso to conjugate perdere and then press the “speak” button (the megaphone) next to the third person plural present indicative from, again it puts the stress on the second syllable.
Therefore, since the Text to Speech algorithms are matching patterns of related written words and associated sounds, and have no concept of the meaning of the sentences, I can see how they would have trouble getting the correct version of perdono in these cases. It then isn’t a great leap to suggest that this might spill over for our Clozemaster voice algorithms into difficulty with perdoni in these sentences.
– Speculation over !! –