3 ms·

It is not just a sell. I build Neural Networks as part of some work I am doing, and the kinds of multi-layer deep networks that do the recognition just don't p

It is not just a sell. I build Neural Networks as part of some work I am doing, and the kinds of multi-layer deep networks that do the recognition just don't perform well on small devices. Also the best systems often have many different trained networks and specialized routing to find the best network for your accent, cadence, noise level, etc. Having all of these weights on your local device, running all the tests, running through any of the required preprocessing, and asking the network what the answer is, converting the results, etc. is going to be too much for your phone or low powered device. While there might be other reasons for them to send data to the cloud, to get the accuracy you really need beefy servers (potentially with GPU's or something more exotic like FPGA's), and potentially multiple servers working in concert.

moron4hire 12y ago
Since I don't know the factors involved and you do, I must then ask how to reconcile this with the fact that the first version of Dragon NaturallySpeaking was released in 1997. My smartphone is definitely more powerful than even a top-of-the-line 1997 desktop PC. Now, I am sure Dragon V1 would be quite underwhelming to my modern notions of how computers should work. But I would even say my smartphone is better than a middle-of-the-road PC from 2010. It's better than my PS2, and even my PS2 had a few games that had speech interface (some Navy Seals game, it worked pretty well). What factors are at play? Is it the noise cancellation? Seems like a stationary device in a home has the ideal chance to create a reusable noise cancellation profile.