The issue is the (many) vocoders along the chain remove anything that don't match the vocal patterns of a human. When you say hello, it's encoded phonetically to a very low bitrate. Noise, or anything outside what a human vocal cord can do, is aggressively filtered or encoded as vocal sounding things. Except for DTMF, which must be preserved for backwards compatibility. That's why I say it would be creepy to do something higher bitrate...your data stream would literally and necessarily be human vocal sounds!