Slashdot Mirror
Google Working On Siri Competitor Majel
judgecorp writes "Google is working on a competitor to Apple's Siri voice input system. It's an extension to its existing Voice Actions offering with a name that should ring bells. Majel is named after Majel Barrett-Roddenberry, who was the voice of most of the Star Trek on-board computers, as well as playing Nurse Christine Chapel in the first series and being Gene Roddenberry's wife."
I'd prefer "Prikazyvat" to "Computer.
I absolutely love the use of "Majel" here; Star Trek has influenced so much of our lives and of our tech, and now that are finally starting to get into responsive voice-operated systems, it shows a great deal of respect to bring it back to the original visionaries.
Aikon-
Google Voice integration is pretty widespread - most of Sprint android users pretty much use it so, yeah, it's widely used.
I don't need a hammer that gets me. I need one I can accurately use. Natural language is very imprecise, a set list of commands makes things more precise.
I find this comment fascinating, and probably helps differentiate geek tools from mass-market tools. Most people prefer accuracy, but I think a lot of geeks really would prefer precision.
But it accurately represents at least a significant part of the Slashdot demographic. I find Siri to be almost completely useless because it isn't designed for precision - it seems to default to a socially acceptable / funny / warm answer. It's often like talking to an Alzheimer's patient - you get a human response, it's just not associated logically with your question. I would much prefer it if Siri could be placed in a 'computer' mode that gave you a more structured syntax.
Other folks, not so much....
Faster! Faster! Faster would be better!
Since I work in SEO I have a good quite understanding how people use search engines. Almost everyone understands how they work and enter good search keywords, just like geeks. On the other hand, sometimes it's just easier to form the keywords as question rather than trying to extract them yourself, and this has the added benefit of similar forum questions coming up first if other people have asked similar questions. It works better with certain kind of queries, so don't label people as "stupid" if they use why, where or what in search queries. I have sometimes needed to research with those words because I couldn't find the information I wanted with keywords, and because of my job I'm quite good at forming them.
The other problem is the whole personal feel. Google is deliberately taking that away with their star trek computer-like interface. I don't even have iPhone, but Apple's Siri seems much more personal. They've made it a character, your friend. Their advertising, features and everything goes along that line. What Google is doing is basically saying "look, we have the technology too!" and forgetting that technology alone isn't enough.
Great Idea Google. I knew exactly who this was named after and why the second i saw the name. Its perfect.
Majel was amazing. TNG for life..
From what I've heard about Siri, it trades accuracy for sophistication of ability and it isn't a good trade off.
All voice recognition systems get around 90%, Siri and Google included. You're only remembering "what you heard" that fits "what your biases confirm". One thing that is very accurate about Siri is when it comes to things on your phone. It will get words that I wouldn't expect any speech recognition system to get (like Dragon), because it gets context. So, for example, if you are talking about music, it will know you said something otherwise non-sequitur, like "Asteroids Galaxy Tour" or "Rush Fly By Night".
I don't know how well Android or MS does with this. From what I've seen on the web, not good at all, but I'm not going to put forth hearsay anecdotes as fact without significant corroboration, even if it does fit what I already believe to be true.
It's really a shame that Majel herself isn't still alive to provide the core voice work for the product. People would have swarmed in droves to have the actual Star Trek computer voice at their beck and call.
Then again, who knows how much audio tape and footage there is of her locked away? Maybe there's enough of a phoneme and phrase collection out there that they could resurrect her voice. Couldn't be any more difficult than extracting the phonemes from someone else's voice, provided there's enough data to do the job.
I do not fail; I succeed at finding out what does not work.
I don't understand what the big deal is. Most of the ideas and algorithms for voice recognition and knowledge processing were already tabled and researched back when I was in 4th year University, in 1986. We just didn't have the compute power to IMPLEMENT anything back then.
It's good to see such technology coming to the forefront, but it's not new ideas. While specifics of the algorithms may be patentable, the concepts pre-date any attempts to patent the ideas, with loads of published research papers and proposals existing as prior art.
Here's a tidbit for you: I first had the idea of inverting a LALR compiler to produce code in 1986 while working on my compiler project for a 400 series class. I worked at it for years, with different tools and technologies, failing time and time again. It wasn't until 1997-1998 that I came up with an approach that was workable, with MSIsa 1.0 in the Java 1.1 era. It took until now to bring it from the conceptual "It can work" implementation to something production worthy for 2012.
But even if I'd patented the idea when I had it, the patent would have expired before I produced a marketable product.
I do not fail; I succeed at finding out what does not work.