That raises an interesting question. Does voice recognition need to be tailored to the type of lossy encoding used? Since perceptual encoders are designed to

That raises an interesting question. Does voice recognition need to be tailored to the type of lossy encoding used? Since perceptual encoders are designed to capture what the human ear thinks it hears, rather than what the sound wave actually looks like, does an algorithm trained on one perceptual encoder apply well to material encoded with another?