1) Yes, it is spun off from research at MILA, University of Montreal. 2) Possibly. Google and Baidu have compute resources far beyond a university. This method

1) Yes, it is spun off from research at MILA, University of Montreal. 2) Possibly. Google and Baidu have compute resources far beyond a university. This method might be better, and do well with more resources. 3) Adobe's method required far more input data. This apparently requires only 1 minute of your audio to start sounding like you. 4) Deep learning has revolutionized vision, and language processing for a while. It was just a matter of time before people started applying those methods on speech data, with similar surprising results. 5) It will be hard to capture human elements like "emotion" and tone via generative models. Maybe in the future the work will become sophisticated enough to be indistinguishable from human speech, but right now there are some telltale signs that it is artificially generated.