For example, Microsoft’s standard model of Spanish will have a default “S” (ess) sound, but the training process replaces it with your “S” sound. This is done f

For example, Microsoft’s standard model of Spanish will have a default “S” (ess) sound, but the training process replaces it with your “S” sound. This is done for every individual sound (phoneme) I don't think it can work as a phoneme for phoneme replacement; there must be a different heuristic at work. Different languages have different phonemes and one of the most telling things of a non-expert non-native speaker is that they don't get the sounds right. Eg "E" is pronounced differently in English, German and Czech. Japanese speakers don't differentiate between "L" and "R". Spanish "C" is in some contexts different to "C"s in other languages. Etc. Obviously you can't pronounce a sound that you don't even know exists. I would love to hear more examples.