The Future of VOIP

Sending voice over the internet is a bit tricky since what you care about is latency (how long from the time a sound is made to when the other person hears it) rather than throughput (how much data you can send over an extended period of time). The internet is largely designed for throughput so people go off and do all sorts of fancy things to improve latency for VOIP (voice over internet protocol).

So if the computer could see a little into the future (like the elevators in hitchhikers guide to the galaxy, or Daniel in the Avatar episode of stargate), over the course of the conversation it would be sending the same amount of data in the same time, but it could deal with short delays. Of course the universe has decided to be causal which ruins so many time travel stories. But the world is increasingly pointing out that human speech really is quite predictable. Type into an iphone and it becomes clear that after just a few letters it often knows what you are going to say. Even more disturbing is the google search box in safari and firefox that will complete what you thought was such a unique search phrase.

So VOIP could see into the future. Obviously it would sometimes be wrong and often would just have a list of possibilities, but it could send across a list of several possibilities (make use of the excess throughput) and then once it knows for sure, just send a short message saying which one it turned out to be. And in the cases where it was wrong then it just sends the voice data as usual. (Has some similarity to pipelining in processors.) You might point out that it doesn't just need to know what word to send, but needs to actually send the sound of your voice, but that just means it has to train over the course of several conversations.

If I was Scott Adams 100 people would now tell me about how this is already being done and another 100 would tell me why it is a stupid idea. Being me, the comment box will sit empty while my friends question if they can delete this RSS feed.