[calancaster]

According to the Video, it also searches your contacts. However when I voice search for someone that I know is in my contacts, the contact info never comes up in my search results.

[doglesby]

Actually the video shows searching for contacts only by typing the name. I'm pretty sure voice search is web-only.

[doglesby]

aryayush said:

Just so it’s clear, I do know that Google does the voice recognition on the server. But there’s a difference between trying to recognise every single word you speak correctly and trying to match what you spoke with the entries in your contacts list. I’m pretty sure the latter is much easier to do than the former and would yield more accurate results.

You might think so, but fundamentally recognizing phonemes is recognizing phonemes. Translating text into phonemes is actually harder with proper names.

People talk about the UI of cut-and-paste because the functionality is easy, the interaction can be done many ways. Speech recognition is the UI, it's the functionality that makes it hard. Notice how little it is used in Mac OS X. I can't even do what you ask on my laptop.

As another example, Say Who uses "powerful speech recognition technology that requires more processing power than the phone can handle." It uploads contact names and searches against them on a server.

I'm pretty sure it's not the interface that is holding voice dialing up.

[VafeR]

My hunch is that the limitations on searching contacts by voice have to do with sending your private data (contact list) out over the network to Google's servers for processing and back - rather than any technical limitation.

If that is the case, I do wonder why they don't just make that clear in a really simple, obvious disclaimer and have the user sign their soul away with a checkbox.

It would be fantastic to use GMA for voice dialing...

(Incidentally, I think this sheds some light on the "why doesn't apple use this tech in their phone" question as well. With server-side-processing, the voice recognition would only work when the user had internet access - which would be annoying. Processing would have to be done on the phone's ARM chip - perhaps it wouldn't be fast enough to do efficiently? I don't know. Does "Say Where" do its processing client-side?

- JR