What is the dominant factor for evaluating performance for TTS & ASR?

Question

We know that CPU speed ("MFLOPS") alone is not the only factor that determines the performance of an Android device.

Other factors include the I/O subsystem, memory speed and, of course, the software that drives this hardware.

But in the case of evaluating various phones in terms of TTS and ASR performance (e.g. minimal delay from the moment you say a word, until it is recognized, given the default built-in engine), can we say that MFLOPS is the dominant factor?

In other words, do you happen to know whether TTS & ASR involve mostly CPU cycles with relatively little memory and I/O transfers? Or is it the opposite?

Answer 1

I would expect TTS and ASR to be very memory-intensive as they process input (which needs to be stored and transformed) and generate output. So slow memory would probably be a limiting factor with a fast CPU. There should also be a little I/O -- loading voice samples for comparison and playback -- but I would not expect it to be as much of an issue. I can't get any more specific since I don't know how the default engine works (I suspect few people outside Google do).

You might get a better answer on StackOverflow, since this is more of a question about the specifics of an algorithm and not really about Android use.