Google has always been a fan of the idea that more data is better, as exemplified by Research Director Peter Norvig’s stance that, generally speaking, more data trumps better algorithms (see, e.g., his 2009 paper titled “The Unreasonable Effectiveness of Data“). Although some hair-splitting does occur about the relative value (or lack thereof) of algorithms in Norvig’s assessment, it’s pretty much an accepted truth at this point and drives much of the discussion around big data. The more data your models have from which to learn, the more accurate they become — even if they weren’t cutting-edge stuff to begin with.

No surprise, then, it turns out that more data is also better for training speech-recognition systems. The researchers found that data sets and larger language models (here’s a Wikipedia explanation of the n-gram type involved in Google’s research) result in fewer errors predicting the next word based on the words that precede it. Discussing the research in a blog post on Wednesday, Google research scientist Ciprian Chelba gives the example that a good model will attribute a higher probability to “pizza” as the next word than to “granola” if the previous two words were “New York.” When it comes to voice search, his team found that “increasing the model size by two orders of magnitude reduces the [word error rate] by 10% relative.”

The real key, however — as any data scientist will tell you — is knowing what type of data is best to train your models, whatever they are. For the voice search tests, the Google researchers used 230 billion words that came from “a random sample of anonymized queries from google.com that did not trigger spelling correction.” However, because people speak and write prose differently than they type searches, the YouTube models were fed data from transcriptions of news broadcasts and large web crawls.

“As far as language modeling is concerned, the variety of topics and speaking styles makes a language model built from a web crawl a very attractive choice,” they write.

This research isn’t necessarily groundbreaking, but helps drive home the reasons that topics such as big data and data science get so much attention these days. As consumers demand ever smarter applications and more frictionless user experiences, every last piece of data and every decision about how to analyze it matters.

Feature image courtesy of Shutterstock user watcharakun.

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• Sector RoadMap: Social customer service in 2013
• A near-term outlook for big data
• Connected world: the consumer technology revolution

The service’s vocabulary and contextual understanding is pretty limited. For instance, if you ask Genius to “find a restaurant” it will pull up the nearest eatery on Google Maps. But if you ask it to find “nearby restaurants” it searches Maps for a joint named “Nearby.” If you’re like me, you’ve probably fiddled with the button a few times and never touched it again, despite its relative convenience on the phone’s faceplate.

But T-Mobile has given Genius a much-needed overhaul, at least on the latest versions of the MyTouch manufactured by Huawei. Nuance Communications, which powers the voice-recognition features on T-Mobile devices, is upgrading Genius’ capabilities and the features of the Genius service, providing a deeper level of natural-language understanding and integrating the service with a much broader array of content sources beyond Maps and Google Search.

For instance, if I were to ask the new Genius for nearby restaurants, it would not only understand my intent but also do its searching on Yelp, pulling up dining options nearby and displaying their rankings and reviews. If I were to change that command to “make a reservation at nearby restaurants,” it would bring me to OpenTable’s website and display eateries in the vicinity that accept online bookings.

If this sounds familiar then you’ve probably used Nuance’s consumer semantic-search app, Dragon Go, available for Android and iOS smartphones. In fact, if you look at the long list of 200 content partners the Genius can access, they’re the same used by Dragon Go. Nuance wouldn’t acknowledge specifically that T-Mobile is white-labeling the semantic-search app, but it’s pretty obvious that’s exactly what it’s doing — and it’s by no means a bad thing. I’m a big fan of Nuance’s intuitive little search app, and being able to access it in fewer steps is a bonus.

The official line, though, is that T-Mobile has basically upgraded its relationship with Nuance to a kind of platinum status. The old Genius tapped into Nuance’s basic speech-recognition APIs, but it had none of the rules-based language-parsing abilities of Nuance’s more-sophisticated offerings. By adding greater contextual understanding and a host of content providers, T-Mobile may be able to turn a pretty lame voice-command feature into something quite useful. So far, though, it’s only available on the MyTouch and MyTouch Q.

Both T-Mobile and Nuance will be represented at GigaOM’s Mobilize conference next month. Nuance CTO Vlad Sejnoha will be speaking on a panel about the future smartphone interface on Sept. 20, while T-Mobile SVP of marketing Brad Duea will discuss the evolution of voice services on Sept. 21.

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• CES 2012: a recap and analysis
• Carrier IQ and the continued erosion of operator trust
• The future of mobile: a segment analysis by GigaOM Pro

AT&T promised in April to open Watson up to general development, but on Monday AT&T SVP for technology and network operations John Donovan made the launch official in a blog post, revealing seven different “contexts,” or language libraries, that app builders can access to speechify their products.

Like Nuance, AT&T has developed specific taxonomies for specific use cases, allowing its servers to anticipate the words and phrasing consumers are likely to speak in a particular context. Those tailored lexicons allow Watson (not to be confused with IBM’s artificial intelligence project Watson) to better capture meaning as well as intent from any given phrase. For instance, the question “where is Trenton?” would be interpreted completely differently if spoken to a Q&A app versus a mapping app.

The contexts are Web Search, Business Search, Voicemail-to-Text transcription, SMS speech transcription, Question and Answer, TV and a category called Generic used for general dictation purposes. The TV context is of particular note because it ties into AT&T’s U-Verse program guide, recognizing queries about particular actors, movies and TV shows.

In general, though, it appears AT&T is only licensing out fairly basic speech recognition and natural language understanding to general developers – features that could be used to build voice search into an app, but nothing that could be used to develop an expansive virtual assistant like Siri.

Watson, however, is likely much more powerful than these contexts let on. Vlingo licenses Watson’s core speech technology for its language servers (though that relationship will likely end after the Nuance acquisition closes), and Research in Motion’s QNX is working with Watson’s API to develop a conceptual connected car assistant. AT&T also has big plans to leverage Watson’s capabilities for its own products. In a recent interview, AT&T emerging devices president Glenn Lurie hinted that Watson could soon make it into its Digital Life connected home technology, becoming the voice-controlled virtual butler we see in science fiction.

Watson does have one big advantage to the average dev: it’s cheap – at least until the end of the year. For the remainder of 2012, AT&T will give free and unlimited access to the Watson APIs to any developer paying AT&T’s $99 annual registration dues. In 2013 a point-based fee structure will kick in. Registered developers will get 5000 points a month and any speech transaction of a minute or less will cost one point. Developers that exceed that amount will be charged $20 for each 2000-point increment, which works out to about 1 cent per transaction.

Image courtesy Flickr user Lazurite.

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• The future of mobile: a segment analysis by GigaOM Pro
• Mobile first-quarter 2013: analysis and outlook
• Analyzing the wearable computing market

Here’s a short promo video on the new Sonar feature that’s clearly taking a fun shot at Apple’s Siri functionality, which launched with the iPhone 4S last October.

As a regular user of the Dolphin Browser — here are eight reasons why — I like the concept of integrated voice actions to use the web. Dolphin has long supported gestures to help navigate the Internet, and the new Sonar feature extends the usefulness of the browser. I’ve been using Sonar for the past day and generally like where the Dolphin team is headed.

You can tap your keyboard microphone button to use Sonar, but I’m a fan of the keyboard-less option: simply shake your handset, and Dolphin Sonar is ready to listen. Obviously, you have to have the browser open when shaking the phone; this functionality isn’t integrated into the operating system.

Most of the time, Sonar simply searches Google for what you need. However, if you want to open a specific website, you can do so through speech. And that’s handy for certain web apps. For example, using Sonar, I could say, “Open weather dot com,” and the mobile site opens immediately, complete with my local forecast. Note that I had previously used the website, so it knew my location. If not, the browser wouldn’t know where I was. Saying “eBay iPad 2″ took me right to the eBay mobile site-results for iPad 2.

I wasn’t able to test the link sharing in Facebook feature as it’s not quite live yet. The concept is great, and I hope the Dolphin folks can extend that functionality to other web services such as Twitter and Google+. All in all, I like what I see in Dolphin Sonar. Some may find it limited, but it’s the type of service that can scale to allow for more functionality in the future as our phones become intelligent assistants.

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• Takeaways from mobile’s second quarter
• A look back at mobile in Q1
• Analyzing the wearable computing market

That’s the promise behind the release of Sensory’s TrulyHandsfree Voice Control 2.0 software Tuesday. The latest speech detection technology builds upon the 1.0 version, which allowed users to trigger a device with a small set of words without a prompt. Now version 2.0 recognizes and responds to many more keywords and longer phrases, dozens of them, that can be analyzed in the course of a regular noisy conversation. The latest update increases the accuracy significantly by 60 percent and picks out words better in loud environments. What it enables is a much more natural interface with a device that can be listening for a wide array of triggers.

TrulyHandsfree is already at work in car Bluetooth kits from BlueAnt and Kensington and phones like the Samsung Galaxy S II. And it’s finding its way into sophisticated toys like Mattel’s Fijit Friends, which can respond to words spoken to it.

A voice-activated world

But that’s just the start. Sensory’s CEO Todd Mozer said he’s been contacted by four TV companies in the last month who are interested in using TrulyHandsfree as a front end for voice controls. He said there could be any number of devices from coffee machines and stereos to home automation systems and automobiles that can utilize the technology. It’s already starting to happen slowly, but now with more sophisticated and accurate software, we should see this integrated into a lot more products.

“Almost any new product can we use this Truly Handsfree approach, it’s so accurate,” Mozer said.

Removing buttons

Mozer said Truly Handsfree is transforming Sensory’s speech technology business. Since launching a two years ago, the product quickly got integrated into 1/3 of all Sensory’s voice-enabled products. This year, 2/3 of all products using Sensory’s technology are utilizing Truly Handsfree.

When combined with smarter speech recognition technology used by companies like Nuance , Vlingo, Google and Microsoft, it could really help bring speech technology into the mainstream. We’re seeing some of that promise now with devices like the Xbox Kinect, which can respond to some voice commands. But with a front end that can respond to a wide array of triggers and work in a lot of different environments, we could see users start to feel comfortable with speech as an input.

Mainstream catalyst

That’s one of things that still holds me up at times. I like Android Voice Actions, but even just the act of pressing the microphone icon to start an action is an extra step that sometimes deters me. Google recently said 25 percent of searches on Android devices are conducted by voice. Imagine how much that would go up if people could just talk to their phones without waking it up. A solution has to be accurate though, not firing on the wrong words. Mozer said Truly Handsfree 2.0 registers a 10 percent false reject rate and has a false acceptance rate of a word every three to five hours, and it can register words from 20 feet away. It’s not perfect, but it is getting very good.

I think speech technology has a ways to go to the point where we’re talking to computers like HAL or the system on the Star Trek Enterprise. But it’s coming together. We still need computers to figure out the harder task of understanding our speech and our intent but that’s coming around as we’re seeing with projects like IBM’s Watson . We increasingly have so much data stored in the cloud and on our devices that it makes sense to also tap tools like voice to get at that. Achieving that future involves a lot of progress including the work being done by Sensory. I, for one, can’t wait to do less button pressing and more voice activating.

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• Bluetooth to Feel Blue as Personal Area Network Battles Loom
• Google doesn’t like walled gardens — except its own
• Netflix may suffer from limited mobility