Post #2388671

@FediThing@social.chinwag.org Speech synthesis based on machine learning sounds much more natural; with a bit of postprocessing, it can't be distinguished from a real human voice anymore, even if no human with that exact voice exists. Also, such AI speech synthesizers are often small enough to run on the local GPU without any need for an external computing centre. There's nothing wrong with using machine learning per se. The problem are big companies trying to make their models bigger and bigger so that they can never be run on consumer hardware.