Lip Reading: A Modest Advance?

November 21, 2016

Watching surveillance videos without sound? Wish you could read lips? Don’t have time to learn how to read lips? Alphabet Google’s DeepMind has a solution for anyone in this predicament. “Google’s DeepMind AI Can Lip-Read TV Shows Better Than a Pro” reveals:

A project by Google’s DeepMind and the University of Oxford applied deep learning to a huge data set of BBC programs to create a lip-reading system that leaves professionals in the dust.

How accurate is the system? The write up states:

The professional annotated just 12.4 per cent of words without any error. But the AI annotated 46.8 per cent of all words in the March to September data set without any error. And many of its mistakes were small slips, like missing an ‘s’ at the end of a word. With these results, the system also outperforms all other automatic lip-reading systems.

I think this means that the Google method is almost four times more accurate. Software is faster and does not require health care, vacation days, and coddling.

The write up sidesteps law enforcement use of the system by emphasizing “improved hearing aids, silent dictation in public spaces, and speech recognition in noisy environments.”

There are other applications, however.

Stephen E Arnold, November 28, 2016

 

HonkinNews for 15 November 2016 Now Available

November 15, 2016

The weekly Beyond Search news video is available at this link. Stories include Mr. Thiel goes to Washington, the “best” entity extraction software and the not-so-best systems. You will learn the latest about the Yahoot security consequences, and more. The video also includes information about the US government’s open source code Web site. Stephen E Arnold points out that the Darpa Dark Web open source code is not included in the Code.gov offerings. Never fear. The Darpa listing does appear in the forthcoming Dark Web Notebook. If you want a copy of this new Beyond Search study, write benkent2020 at yahoo dot com and reserve your password protected PDF today.

Over the New Year’s break three, free special seven minute programs will air on December 20, December 27, and January 3, 2017, HonkinNews will run a weekly seven minute video. Each video presents the principal takeaways from Stephen E Arnold’s Google Trilogy: The Google Legacy (2004), Google Version 2 (2007), and Google: The Digital Gutenberg (2009). The information remains timely even though Alphabet Google is in a somewhat excited state of shifting in order to generate revenue as the volume of searches from the desktop declines squishing Google’s online ad methods for old fashioned Internet access.

Kenny Toth, November 15, 2016

HonkinNews for November 8, 2016 Now Available

November 8, 2016

This week HonkinNews comments about Microsoft’s mobile phone adventure. You will learn about geo spatial analytics’ companies that may have an impact in certain secret applications. Palantir  makes news again. There is more. You can view the seven minute video at this link https://youtu.be/UWCk4n_AC0Y.

Kenny Toth, November 8, 2016

MicroSearch: A Specialized Video Search System for Academic Content

November 3, 2016

I am not sure if the Alphabet Google thing will be down with this new video search system over the long haul. If you want a different way to locate academic videos, you will want to explore MicroSearch’s system. MicroSearch says that it is “a boutique search engine company, providing private, secure video and document cloud storage as well as custom search services.”

I learned about this service in “University Videos on YouTube Get Custom Search.” You can explore the system at http://universityvideos.org/Home. The search system is at this link.

image

According to the write up, the system aggregates university videos and:

includes a video player that shows the video playing on the left and a transcript tracking with the video on the right. Clicking into another sentence in the transcript jumps the user to that part of the video.

I highlighted this passage:

The service also includes a search tool that allows the user to search on transcript contents, title, description, duration, category, tags, YouTube channel and year uploaded. The same fields are available as metadata, when search results are displayed and downloaded as an Excel export file. An advanced search feature lets the user enter a few letters into the transcription field and then click on an Index button next to the field to obtain a window that displays all of the terms with that series of letters.

Our test queries suggested that the system is less wonky than Google’s video search. The fact that Google is splitting its text index into one part for mobile and one part for traditional desktop search makes clear that search at Google is a work in progress. With a new search system for a segment of YouTube videos, one can conclude that YouTube video search is not a home run for some users.

Perhaps more attention on search and less on Loon balloons might solve the problem. On the other hand, Alphabet Google can simply block developers of “better mousetraps” and move forward with its online advertising programs and projects like solving death. Search is for revenue and maybe not for finding relevant content?

Stephen E Arnold, November 3, 2016

HonkinNews, 1 November 2016 Now Available

November 1, 2016

This week’s HonkinNews explores the mini Silicon Valley the White House wanted to build on the banks of the swampy Potomac River. How did the venture perform? Think F as in 18F. The British equivalent of the GSA is not winning high fives among some UK citizens. The IBM WOW showed Las Vegas how to party with Big Blue professionals, Watson (a true party animal), and true believers. Do you have a stack of the new currency? IBM believes it has the combination to unlock a vault containing riches. Google makes it more difficult to find out which Google wizard is responsible for what. HonkinNews provides some tips to locate the Googler you want to reach. Attivo breaks new ground for verbal finery. Verizon believes that it has the secret to unlock the value of Yahoo. AT&T, it seems, is heading into trouble. Click here for the program.

Kenny Toth, November 1, 2016

Microsoft Can Understand Human Conversation Like a Human

October 31, 2016

I read “Microsoft Speech Recognition Technology Now Understands a Conversation As Well As a Person.” My wife’s Amazon Alexa does okay with her commands. I noted this passage in the write up:

This marks the first time that human parity has been reported for conversational speech.

Okay, I will inform my wife that Alexa is not able to do the speech recognition thing. She gave up on Microsoft Windows, laughed at the Windows phone I gave her, and bought a Mac laptop. She seems okay with what her iPhone 6 can do, but I will try again to explain that Microsoft really, really has solved a hard problem.

The write up points out:

In a paper published this week the Microsoft Artificial Intelligence and Research group said its speech recognition system had attained “human parity” and made fewer errors than a human professional transcriptionist.

Oh, not a product or a service she can test yet. The innovation is embodied in a paper. Is this content marketing or public relations? I suppose I could ask Cortana if we had a machine running that particular Microsoft invention. Windows 10 left us some time ago. Sorry.

The error rate of about six percent seems okay until you think about six words in 100 being incorrect. Some situatio0ns pivot on a single word, don’t they?

I will wait for the new system to be hooked up to Microsoft Tay. I remember Tay. The system learned some of the less savory aspects of language before the demonstration was sent back to the lab. The interaction of speech recognition and Tay will be something I want to test.

Maybe my wife will have a change of heart with regards to Apple and Amazon products.

Stephen E Arnold, October 31, 2016

HonkinNews for October 18, 2016 Now Available

October 18, 2016

From the wilds of rural Kentucky, Stephen E Arnold highlights the week’s search, online, and content processing news. Two services make it easy to buy a product with a mouse click. Will Amazon’s eCommerce business be threatened by eBay and Pinterest? Plus, this week’s program comments about Google and Pindrop, National Geographic’s new topographic maps, and another of Yahoo’s mounting public relations challenges. The program explains that Google is taking a step toward marginalizing the “regular” Web in favor of the mobile Web. You can view the video shot in eight millimeter film from a cabin in a hollow at this link.

Kenny Toth, October 18, 2016

eBay and Corrigon: Heading in the Right Direction?

October 14, 2016

I find eBay fascinating. Many things for sale; for example, $3,000 Teddy bears. I wonder what those are.

I read “eBay to Acquire Corrigon Ltd.” Interesting. I learned about Corrigon, an Israel-based image search and analysis outfit, about seven years ago. The company’s technology can “look” at a digital image and recognize objects in the image. Coirrigon’s pitch, as I recall it, introduced me to the concept of “dynamic browsing.” I thought most browsing was, by definition, was dynamic, but why ask questions which marketers cannot or will not answer. The buzzwords are the intellectual food which gives me Delhi belly.

One application of Corrigon’s technology is to identify objects in a photo can create a link to a shopping site where one can purchase that object. For instance, I am looking at this image:

image

The Corrigon system will, in theory, point me to this type of entry on another Web site:

image

What if I really want the model’s shirt? Well, that may be an issue.

Corrigon has some law enforcement and intelligence applications as well. My hunch is that eBay wants to allow a person to see something, buy something.

The method adds layers and performs image parsing. The method is fine but the approach can add compute cycles. Latency when shopping is a bit of brown bread.

The write up informed me that:

Corrigon’s technology and expertise will contribute to eBay’s efforts with image recognition, classification and image enhancements as part of its structured data initiative. There are three parts to eBay’s structured data initiative: first, collect the data; second, process and enrich the data; and third, create product experiences.Corrigon will support the second and third parts – processing and enriching the data and creating product experiences.

Let’s think about how an eBay user accesses information in the digital flea market now. A person navigates to the site and plugs in keywords. The system then generates a bewildering array of options and some listings. A user then scans and clicks the laundry list of listing. Then the user reads individual listings. Then the user presumably buys the best listing. Heaven help the user who needs to hunt for the link to ask the seller a question. Etc. etc. etc.

eBay’s purchase of Corrigon is going to make eBay into a zippier shopping experience. Well, that’s the theory.

eBay’s challenge is my fave Craigslist and obviously the Bezos beastie. I asked myself, “Perhaps eBay should do some interface work and poke around its core search functionality?”

Stephen E Arnold, October 14, 2016

HonkinNews for October 11, 2016 Now Available

October 10, 2016

The most recent HonkinNews video is now available at this link. Stories include Yahoo’s most recent adventure: A purple light Y-Mart discount of $1 billion dollars on the Verizon purchase offer. Learn how Google Translate handles a Chinese poem about ospreys, not government administration. Included in the seven minute program is information about IBM Watson in the third grade and Bing’s secret to revenue success. These stories and more like the diffusion of the idea of “good enough” search. Direct from Harrod’s Creek in rural Kentucky… HonkinNews for the week ending October 11, 2016.

Stephen E Arnold, October 10, 2016

HonkinNews for October 4, 2016 Available

October 4, 2016

This week’s HonkinNews is available at this link. The feature story explores Palantir Technologies’ love-less love relationship with the US Army. Palantir’s approach to keeping its government customers happy is innovative. We also comment about Google’s blurring of cow faces in StreetView. Learn why SearchBlox is giving vendors of expensive, proprietary enterprise search systems cramps in their calves. Microsoft continues to pay users to access the Internet via Edge and use Bing to search for information. How much does the US government spend for operations and maintenance of its systems? The figure is surprising, if not shocking. This and more in HonkinNews for October 4, 2016.

Kenny Toth, October 4, 2016

« Previous PageNext Page »