Mazty said:
timmah said:
Mazty said:
timmah said:
You must have missed what I said about picking out keywords from the sentence. This greatly simplifies the overall process and would allow something like what I'm describing to work. For the example we're using, no matter how you ask to upgrade your suit, the words 'upgrade' and 'suit' should always be in that sentence (hell, even if only the word 'suit' was in the sentence, it could say 'What would you like to do with your suit' and give you options), so the system knows which menu to take you into based on those keywords alone. From there it would be as simple as the computer asking you which upgrade you would like to apply, and either inputting the answer via touchscreen or voice. Because this is in-game and therefore very specific to that game, and because you're interacting with a virtual assistant that only performs specific functions within the game, the number of commands does not have to be even close to what Siri is able to process. It's been a while since I've done any programming, but when I was in college (back in '02) one of my classmates made a proof of concept in VB similar to this and was able to process queries, then present a response from natural sentence structure by picking out pre-programmed keywords without having to parse the entire sentence using either keyboard or off-the-shelf voice recognition software as the input, and without using significant processing power... so I know it could be done. It probably won't, but it is possible.
Though something like this could (unfairly) get compared to Siri, there could still be some cool applications.
EDIT: Though if this were implemented via the cloud like Siri, this discussion would become a moot point.
|
But you still have to record the entire sentence to determine where the key words are. You can't just "go deaf" when it's not a word that is recognised. I don't understand how you couldn't parse the entire sentence as the entire sentence still needs to be analysed to find the key words.
The only issue with the cloud is latency, although if Nintendo managed the cloud, that would seem more realistic then a dev or publisher.
|
If you convert the spoken sentence to text, the system does not have to understand the entire sentence. It can analyze the words as text and process only the ones in a keywords database, resulting in a more simple command to execute, such as turning the phrase "Let's go ahead and upgrade my suit" into "upgrade suit". It is much simpler to extract keywords than to process and understand an entire sentence using natural speech and grammar, thus requiring significantly less complex code, making for significantly less processing power required.
|
I know it doesn't have to understand the sentence, but it still has to record the entire setence and convert it. PC software that does just that takes a dual core a@ 2.2Ghz if we go by retail software.
I understand exactly what you are saying, but I think you aren't realising that it still needs to be able to be write down correctly everything that is written.
|
The part of 'writing down what is said' is a very small part of the power equation, and has been possible at some level since the late 80's. Dragon takes a bit more power because it has to take natural speech and know where to insert punctuation, make paragraph breaks, etc. meaning it needs to understand complex language structure and create properly formatted sentences. That being said, it still doesn't take a very large chunk out of overall processor time, AND, the first versions of the software came out in the 90's and worked on those terrible processors. In addition, there were a lot of applications that did speech to text in the 90's with very, very limited processing power. You're flat wrong in thinking translating speech to simple text takes much power at all. It just doesn't.
EDIT: By the way, smartphones actually do the voice recognition on the device (using a low powered ARM processor), then upload the translated text to the search service (such as Siri or Google Voice Actions) for processing of the command. You can prove this on an android phone by enabling airplane mode, then using the speech to text feature in the compose email window. It processes the speech to text without any access to the web, meaning on the local device.
