16 August 2019

Joint Speech Recognition and Speaker Diarization via Sequence Transduction




Being able to recognize “who said what,” or speaker diarization, is a critical step in understanding audio of human dialog through automated means. For instance, in a medical conversation between doctors and patients, “Yes” uttered by a patient in response to “Have you been taking your heart medications regularly?” has a substantially different implication than a rhetorical “Yes?” from a physician.

Conventional speaker diarization (SD) systems use two stages, the first of which detects changes in the acoustic spectrum to determine when the speakers in a conversation change, and the second of which identifies individual speakers across the conversation. This basic multi-stage approach is almost two decades old, and during that time only the speaker change detection component has improved.

With the recent development of a novel neural network model—the recurrent neural network transducer (RNN-T)—we now have a suitable architecture to improve the performance of speaker diarization addressing some of the limitations of the previous diarization system we presented recently. As reported in our recent paper, “Joint Speech Recognition and Speaker Diarization via Sequence Transduction,” to be presented at Interspeech 2019, we have developed an RNN-T based speaker diarization system and have demonstrated a breakthrough in performance from about 20% to 2% in word diarization error rate—a factor of 10 improvement.

Conventional Speaker Diarization Systems
Conventional speaker diarization systems rely on differences in how people sound acoustically to distinguish the speakers in the conversations. While male and female speakers can be identified relatively easily from their pitch using simple acoustic models (e.g., Gaussian mixture models) in a single stage, speaker diarization systems use a multi-stage approach to distinguish between speakers having potentially similar pitch. First, a change detection algorithm breaks up the conversation into homogeneous segments, hopefully containing only a single speaker, based upon detected vocal characteristics. Then, deep learning models are employed to map segments from each speaker to an embedding vector. Finally, in a clustering stage, these embeddings are grouped together to keep track of the same speaker across the conversation.

In practice, the speaker diarization system runs in parallel to the automatic speech recognition (ASR) system and the outputs of the two systems are combined to attribute speaker labels to the recognized words.
Conventional speaker diarization system infers speaker labels in the acoustic domain and then overlays the speaker labels on the words generated by a separate ASR system.
There are several limitations with this approach that have hindered progress in this field. First, the conversation needs to be broken up into segments that only contain speech from one speaker. Otherwise, the embedding will not accurately represent the speaker. In practice, however, the change detection algorithm is imperfect, resulting in segments that may contain multiple speakers. Second, the clustering stage requires that the number of speakers be known and is particularly sensitive to the accuracy of this input. Third, the system needs to make a very difficult trade-off between the segment size over which the voice signatures are estimated and the desired model accuracy. The longer the segment, the better the quality of the voice signature, since the model has more information about the speaker. This comes at the risk of attributing short interjections to the wrong speaker, which could have very high consequences, for example, in the context of processing a clinical or financial conversation where affirmation or negation needs to be tracked accurately. Finally, conventional speaker diarization systems do not have an easy mechanism to take advantage of linguistic cues that are particularly prominent in many natural conversations. An utterance, such as “How often have you been taking the medication?” in a clinical conversation is most likely uttered by a medical provider, not a patient. Likewise, the utterance, “When should we turn in the homework?” is most likely uttered by a student, not a teacher. Linguistic cues also signal high probability of changes in speaker turns, for example, after a question.

There are a few exceptions to the conventional speaker diarization system, but one such exception was reported in our recent blog post. In that work, the hidden states of the recurrent neural network (RNN) tracked the speakers, circumventing the weakness of the clustering stage. Our approach takes a different approach and incorporates linguistic cues, as well.

An Integrated Speech Recognition and Speaker Diarization System
We developed a novel and simple model that not only combines acoustic and linguistic cues seamlessly, but also combines speaker diarization and speech recognition into one system. The integrated model does not degrade the speech recognition performance significantly compared to an equivalent recognition only system.

The key insight in our work was to recognize that the RNN-T architecture is well-suited to integrate acoustic and linguistic cues. The RNN-T model consists of three different networks: (1) a transcription network (or encoder) that maps the acoustic frames to a latent representation, (2) a prediction network that predicts the next target label given the previous target labels, and (3) a joint network that combines the output of the previous two networks and generates a probability distribution over the set of output labels at that time step. Note, there is a feedback loop in the architecture (diagram below) where previously recognized words are fed back as input, and this allows the RNN-T model to incorporate linguistic cues, such as the end of a question.
An integrated speech recognition and speaker diarization system where the system jointly infers who spoke when and what.
Training the RNN-T model on accelerators like graphical processing units (GPU) or tensor processing units (TPU) is non-trivial as computation of the loss function requires running the forward-backward algorithm, which includes all possible alignments of the input and the output sequences. This issue was addressed recently in a TPU friendly implementation of the forward-backward algorithm, which recasts the problem as a sequence of matrix multiplications. We also took advantage of an efficient implementation of the RNN-T loss in TensorFlow that allowed quick iterations of model development and trained a very deep network.

The integrated model can be trained just like a speech recognition system. The reference transcripts for training contain words spoken by a speaker followed by a tag that defines the role of the speaker. For example, “When is the homework due?” ≺student≻, “I expect you to turn them in tomorrow before class,” ≺teacher≻. Once the model is trained with examples of audio and corresponding reference transcripts, a user can feed in the recording of the conversation and expect to see an output in a similar form. Our analyses show that improvements from the RNN-T system impact all categories of errors, including short speaker turns, splitting at the word boundaries, incorrect speaker assignment in the presence of overlapping speech, and poor audio quality. Moreover, the RNN-T system exhibited consistent performance across conversation with substantially lower variance in average error rate per conversation compared to the conventional system.

A comparison of errors committed by the conventional system vs. the RNN-T system, as categorized by human annotators.
Furthermore, this integrated model can predict other labels necessary for generating more reader-friendly ASR transcripts. For example, we have been able to successfully improve our transcripts with punctuation and capitalization symbols using the appropriately matched training data. Our outputs have lower punctuation and capitalization errors than our previous models that were separately trained and added as a post-processing step after ASR.

This model has now become a standard component in our project on understanding medical conversations and is also being adopted more widely in our non-medical speech services.

Acknowledgements
We would like to thank Hagen Soltau without whose contributions this work would not have been possible. This work was performed in collaboration with Google Brain and Speech teams.

YC-backed Lokal wants to bring local news, classifieds to 900 million Indians in their regional languages


Each month millions of Indians are coming online for the first time, making India the last great growth market for internet companies worldwide. But winning them presents its own challenges.

These users, most of whom live in small cities and villages in India, can’t speak English. Their interests and needs are different from those of their counterparts in large cities. When they come online, the world wide web that is predominantly focused on the English-speaking masses, suddenly seems tiny, Google executives acknowledged at a media conference last year. According to a KPMG-Google report (PDF) on Indian languages, there will be 536 million non-English speaking users using internet in India by 2021.

Many companies are increasingly adding support for more languages, and Silicon Valley giants such as Google are developing tools to populate the web with content in Indian languages.

But there is still room for others to participate. On Friday, a new startup announced it is also in the race. And it has already received the backing of Y Combinator (YC).

Lokal is a news app that wants to bring local news to hundreds of millions of users in India in their regional languages. The startup, which is currently available in the Telugu language, has already amassed more than two million users, Jani Pasha, co-founder of Lokal, told TechCrunch in an interview.

lokal homescreen

There are tens of thousands of publications in India and several news aggregators that showcase the top stories from the mainstream outlets. But very few today are focusing on local news and delivering it in a language that the masses can understand, Pasha said.

Lokal is building a network of stringers and freelance reporters who produce original reporting around the issues and current affairs of local towns and cities. The app is updated throughout the day with regional news and also includes an “information” stream that shows things like current price of vegetables, upcoming events and contact details for local doctors and police stations.

The platform has grown to cover 18 districts in South India and is slowly ramping up its operations to more corners of the country. The early signs show that people are increasingly finding Lokal useful. “In 11 of the 18 districts we cover, we already have a larger presence and reader base than other media houses,” Pasha said.

Before creating Lokal, Pasha and the other co-founder of the startup, Vipul Chaudhary, attempted to develop a news aggregator app. The app presented news events in a timeline, offering context around each development.

“We made the biggest mistake. We built the product for four to five months without ever consulting with the users. We quickly found that nobody was using it. We went back to the drawing board and started interviewing users to understand what they wanted. How they consumed news, and where they got their news from,” he said.

“One thing we learned was that most of these users in tier 2 and tier 3 India still heavily rely on newspapers. Newspapers still carry a lot of local news and they rely on stringers who produce these news pieces and source them to publications,” he added.

But newspapers have limited pages, and they are slow. So Pasha and the team tried to build a platform that addresses these two things.

Pasha tried to replicate it through distributing local news, sourced from stringers, on a WhatsApp group. “That one WhatsApp group quickly became one of many as more and more people kept joining us,” he recalls. And that led to the creation of Lokal.

Along the journey, the team found that classifieds, matrimonial ads and things like birthday wishes are still driving people to newspapers, so Lokal has brought those things to the platform.

Pasha said Lokal will expand to three more states in the coming months. It will also begin to experiment with monetization, though that is not the primary focus currently. “The plan is to eventually bring this to entire India,” he said.

A growing number of startups today are attempting to build solutions for what they call India 2 and India 3 — the users who don’t live in major cities, don’t speak English and are financially not as strong.

ShareChat, a social media platform that serves users in 15 regional languages — but not English — said recently it has raised $100 million in a round led by Twitter. The app serves more than 60 million users each month, a figure it wants to double in the next year.


Read Full Article

Twitter to test a new filter for spam and abuse in the Direct Message inbox


Twitter is testing a new way to filter unwanted messages from your Direct Message inbox. Today, Twitter allows users to set their Direct Message inbox as being open to receiving messages from anyone, but this can invite a lot of unwanted messages, including abuse. While one solution is to adjust your settings so only those you follow can send your private messages, that doesn’t work for everyone. Some people — like reporters, for example — want to have an open inbox in order to have private conversations and receive tips.

This new experiment will test a filter that will move unwanted messages, including those with offensive content or spam, to a separate tab.

Instead of lumping all your messages into a single view, the Message Requests section will include the messages from people you don’t follow, and below that, you’ll find a way to access these newly filtered messages.

Users would have to click on the “Show” button to even read these, which protects them from having to face the stream of unwanted content that can pour in at times when the inbox is left open.

And even upon viewing this list of filtered messages, all the content itself isn’t immediately visible.

In the case that Twitter identifies content that’s potentially offensive, the message preview will say the message is hidden because it may contain offensive content. That way, users can decide if they want to open the message itself or just click the delete button to trash it.

The change could allow Direct Messages to become a more useful tool for those who prefer an open inbox, as well as an additional means of clamping down on online abuse.

It’s also similar to how Facebook Messenger handles requests — those from people you aren’t friends with are relocated to a separate Message Requests area. And those that are spammy or more questionable are in a hard-to-find Filtered section below that.

It’s not clear why a feature like this really requires a “test,” however — arguably, most people would want junk and abuse filtered out. And those who for some reason did not, could just toggle a setting to turn the filter off.

Instead, this feels like another example of Twitter’s slow pace when it comes to making changes to clamp down on abuse. Facebook Messenger has been filtering messages in this way since late 2017. Twitter should just launch a change like this, instead of “testing” it.

The idea of hiding — instead of entirely deleting — unwanted content is something Twitter has been testing in other areas, too. Last month, for example, it began piloting a new “Hide Replies” feature in Canada, which allows users to hide unwanted replies to their tweets so they’re not visible to everyone. The tweets aren’t deleted, but rather placed behind an extra click — similar to this Direct Message change.

Twitter is updating is Direct Message system in other ways, too.

At a press conference this week, Twitter announced several changes coming to its platform including a way to follow topics, plus a search tool for the Direct Message inbox, as well as support for iOS Live Photos as GIFs, the ability to reorder photos, and more.


Read Full Article

Motorola has GoPro aspirations for its latest budget handset


Good on Motorola for keeping it interesting. At the end of the day, I’m not sure how large a market there is for a budget “action camera” phone, but in a world of samey electronics, at least the Lenovo-owned brand continues to shade in interesting corners.

Honestly, I’m surprised more phones didn’t attempt to position themselves as action devices in the heyday of the GoPro. For most consumers, that trend has largely blown over, and for GoPro itself, it’s become tough to compete with cheap knockoffs and the recent entrance of DJI into the market.

The “action” part of the Motorola One Action is largely a reference to the three camera array on the rear — specifically the 117 degree, ultra side angle that offers a more GoPro-style shot. What’s really interesting here (and could have broad implications) is the decision to change the sensor’s orientation. Holding the phone vertically (wrong) results in landscape videos (right).

And listen, I realize that there’s no “right” or “wrong” way to shoot phone video, but come on, let’s be real for a minute. Those bars a really awful way to watch a YouTube video.

The handset is pretty much a standard budget Moto device in ever other way. It arrives in Brazil, Mexico and parts of Europe today. The U.S. and Canada will get their hands on it in October. Expect it to be priced under $300.


Read Full Article

Serena Versus the Drones


Serena Versus the Drones

The Portable RetroStone 2 Emulates Dozens of Retro Gaming Consoles


What’s old is new again. At least, that seems to be the case in the world of video games, where retro video games are more popular than ever.

A company at the forefront of this retro gaming revolution is 8BCraft. It has already successfully launched two classic game products in the Raspiboy and RetroStone, and now the team is back again with the RetroStone 2. For anyone who loves playing retro games, this portable console is exactly what you’ve been looking for.

What Is the RetroStone 2?

The RetroStone 2 takes full advantage of the power of Retropie. Obviously, that means you’re going to downloading and running ROMs and emulators in order to play games. Make sure you’re aware of the laws for downloading and using ROMs in your area, as they vary from country to country.

Because the console is designed to be played on the go, it comes with a built-in 3.5-inch screen. But what makes it really cool is the HDMI port, which means you easily plug your console into the big screen and play your favorite classics. There’s also an ethernet port, three USB ports, and a microSD card slot.

It also has all of the necessary buttons to play your favorite games from a wide range of consoles. On the front, there’s a d-pad, four face buttons, a start, and a select button. On the back of the console are four additional buttons that’ll be necessary for games on consoles like the SNES, PS1, and any newer system.

As for the internals, there’s a single-board computer that features an A20 processor with two cores pushing 1.0GHz of power. There’s also 1GB of RAM. There are stretch goals for Wi-Fi and Bluetooth at €100,000 and €150,000, respectively.

If you spend the extra money on a Pro version, you’ll get 8GB of built-in memory, a SATA m.2 connector, a joystick connector, and a 64GB SD card.

Here are some other things to note about the RetroStone 2:

  • Removable 4000 mAh battery
  • 640X480 resolution display
  • Built-in speaker
  • 3.5mm headphone jack

RetroStone 2 Price and Availability

8BCraft is seeking funding for its RetroStone 2 on Kickstarter. It has vastly exceeded its funding goal already, so the old-school games console should find its way to market.

If you’re interested in preordering the console, you can do so for €129 (about US$144). The company expects to deliver the devices in December 2019, so you won’t have to wait too long to get your hands on one.

Even though this particular company has already delivered products to buyers, there are still risks involved in backing any crowdfunding campaign, and you should make yourself aware of these before you spend your hard-earned cash.

Read the full article: The Portable RetroStone 2 Emulates Dozens of Retro Gaming Consoles


Read Full Article

Twitter leads $100M round in top Indian regional social media platform ShareChat


Is there room for another social media platform? ShareChat, a four-year-old social network in India that serves tens of million of people in regional languages, just answered that question with a $100 million financing round led by global giant Twitter.

Other than Twitter, TrustBridge Partners, and existing investors Shunwei Capital, Lightspeed Venture Partners, SAIF Capital, India Quotient and Morningside Venture Capital also participated in the Series D round of ShareChat.

The new round, which pushes ShareChat’s all-time raise to $224 million, valued the firm at about $650 million, a person familiar with the matter told TechCrunch. ShareChat declined to comment on the valuation.

sharechat screenshot

Screenshot of Sharechat home page on web

“Twitter and ShareChat are aligned on the broader purpose of serving the public conversation, helping the world learn faster and solve common challenges. This investment will help ShareChat grow and provide the company’s management team access to Twitter’s executives as thought partners,” said Manish Maheshwari, managing director of Twitter India, in a prepared statement.

Twitter, like many other Silicon Valley firms, counts India as one of its key markets. And like Twitter, other Silicon Valley firms are also increasingly investing in Indian startups.

ShareChat serves 60 million users each month in 15 regional languages, Ankush Sachdeva, co-founder and CEO of the firm, told TechCrunch in an interview. The platform currently does not support English, and has no plans to change that, Sachdeva said.

That choice is what has driven users to ShareChat, he explained. The early incarnation of the social media platform supported English language. It saw most of its users choose English as their preferred language, but this also led to another interesting development: Their engagement with the app significantly reduced.

The origin story

“For some reason, everyone wanted to converse in English. There was an inherent bias to pick English even when they did not know it.” (Only about 10% of India’s 1.3 billion people speak English. Hindi, a regional language, on the other hand, is spoken by about half a billion people, according to official government figures.)

So ShareChat pulled support for English. Today, an average user spends 22 minutes on the app each day, Sachdeva said. The learning in the early days to remove English is just one of the many things that has shaped ShareChat to what it is today and led to its growth.

In 2014, Sachdeva and two of his friends — Bhanu Singh and Farid Ahsan, all of whom met at the prestigious institute IIT Kanpur — got the idea of building a debate platform by looking at the kind of discussions people were having on Facebook groups.

They identified that cricket and movie stars were popular conversation topics, so they created WhatsApp groups and aggressively posted links to those groups on Facebook to attract users.

It was then when they built chatbots to allow users to discover different genres of jokes, recommendations for phones and food recipes, among other things. But they soon realized that users weren’t interested in most of such offerings.

“Nobody cared about our smartphone recommendations. All they wanted was to download wallpapers, ringtones, copy jokes and move on. They just wanted content.”

sharechat team

So in 2015, Sachdeva and company moved on from chatbots and created an app where users can easily produce, discover and share content in the languages they understand. (Today, user generated content is one of the key attractions of the platform, with about 15% of its user base actively producing content.)

A year later, ShareChat, like tens of thousands of other businesses, was in for a pleasant surprise. India’s richest man, Mukesh Ambani, launched his new telecom network Reliance Jio, which offered users access to the bulk of data at little to no charge for an extended period of time.

This immediately changed the way millions of people in the country, who once cared about each megabyte they consumed online, interacted with the internet. On ShareChat people quickly started to move from sharing jokes and other messages in text format to images and then videos.

Path ahead and monetization

That momentum continues to today. ShareChat now plans to give users more incentive — including money — and tools to produce content on the platform to drive engagement. “There remains a huge hunger for content in vernacular languages,” Sachdeva said.

Speaking of money, ShareChat has experimented with ads on the app and its site, but revenue generation isn’t currently its primary focus, Sachdeva said. “We’re in the Series D now so there is obviously an obligation we have to our investors to make money. But we all believe that we need to focus on growth at this stage,” he said.

ShareChat, which is headquartered in Bangalore, also has many users in Bangladesh, Nepal and the Middle East, where many users speak Indian regional languages. But the startup currently plans to focus largely on expanding its user base in India, hopefully doubling it in the next one year, he said.

It will use the new capital to strengthen the technology infrastructure and hire more tech talent. Sachdeva said ShareChat is looking to open an office in San Francisco to hire local engineers there.

A handful of local and global giants have emerged in India in recent years to cater to people in small cities and villages, who are just getting online. Pratilipi, a storytelling platform has amassed more than 5 million users, for instance. It recently raised $15 million to expand its user base and help users strike deals with content studios.

Perhaps no other app poses a bigger challenge to ShareChat than TikTok, an app where users share short-form videos. TikTok, owned by one of the world’s most valued startups, has over 120 million users in India and sees content in many Indian languages.

But the app — with its ever growing ambitions — also tends to land itself in hot water in India every few weeks. In all sensitive corners of the country. On that front, ShareChat has an advantage. Over the years, it has emerged as an outlier in the country that has strongly supported proposed laws by the Indian government that seek to make social apps more accountable for content that circulates on their platforms.


Read Full Article

Toolkit for digital abuse could help victims protect themselves


Domestic abuse comes in digital forms as well as physical and emotional, but a lack of tools to address this kind of behavior leaves many victims unprotected and desperate for help. This Cornell project aims to define and detect digital abuse in a systematic way.

Digital abuse may be many things: hacking the victim’s computer, using knowledge of passwords or personal date to impersonate them or interfere with their presence online, accessing photos to track their location, and so on. As with other forms of abuse, there are as many patterns as there are people who suffer from it.

But with something like emotional abuse, there are decades of studies and clinical approaches to address how to categorize and cope with it. Not so with newer phenomena like being hacked or stalked via social media. That means there’s little standard playbook for them, and both abused and those helping them are left scrambling for answers.

“Prior to this work, people were reporting that the abusers were very sophisticated hackers, and clients were receiving inconsistent advice. Some people were saying, ‘Throw your device out.’ Other people were saying, ‘Delete the app.’ But there wasn’t a clear understanding of how this abuse was happening and why it was happening,” explained Diana Freed, a doctoral student at Cornell Tech and co-author of a new paper about digital abuse.

“They were making their best efforts, but there was no uniform way to address this,” said co-author Sam Havron. “They were using Google to try to help clients with their abuse situations.”

Investigating this problem with the help of a National Science Foundation grant to examine the role of tech in domestic abuse, they and some professor collaborators at Cornell and NYU came up with a new approach.

There’s a standardized questionnaire to characterize the type of tech-based being experienced. It may not occur to someone who isn’t tech-savvy that their partner may know their passwords, or that there are social media settings they can use to prevent that partner from seeing their posts. This information and other data are added to a sort of digital presence diagram the team calls the “technograph” and which helps the victim visualize their technological assets and exposure.

technograph filled

The team also created a device they call the IPV Spyware Discovery, or ISDi. It’s basically spyware scanning software loaded on a device that can check the victim’s device without having to install anything. This is important because an abuser may have installed tracking software that would alert them if the victim is trying to remove it. Sound extreme? Not to people fighting a custody battle who can’t seem to escape the all-seeing eye of an abusive ex. And these spying tools are readily available for purchase.

“It’s consistent, it’s data-driven and it takes into account at each phase what the abuser will know if the client makes changes. This is giving people a more accurate way to make decisions and providing them with a comprehensive understanding of how things are happening,” explained Freed.

Even if the abuse can’t be instantly counteracted, it can be helpful simply to understand it and know that there are some steps that can be taken to help.

The authors have been piloting their work at New York’s Family Justice Centers, and following some testing have released the complete set of documents and tools for anyone to use.

This isn’t the team’s first piece of work on the topic — you can read their other papers and learn more about their ongoing research at the Intimate Partner Violence Tech Research program site.


Read Full Article

How to Spot Unsafe Email Attachments: 6 Red Flags


disposable-email

Email remains a prominent attack vector for hackers, cybercriminals, snoopers, and other online miscreants. As such, it’s vital that you know how to spot an unsafe email attachment.

If you’re not sure where to start, keep reading. We’re going to explain several red flags that’ll help you identify potentially dangerous files in your inbox.

1. Dangerous File Extensions

Unfortunately, there are several file extensions which could potentially run code on your computer and thus install malware.

As you’d expect, hackers don’t make them easy to spot. Often, dangerous file extensions are concealed in ZIP files and RAR archives. If you see either of those extensions in an attachment that doesn’t come from a recognized contact, you should treat it with suspicion.

The most dangerous file extension is EXE. They are Windows executable files which are particularly hazardous due to their ability to disable your antivirus app.

Other frequently used extensions to watch out for include:

  • JAR: They can take advantage of Java runtime insecurities.
  • BAT: Contains a list of commands that run in MS-DOS.
  • PSC1: A PowerShell script with commands.
  • VB and VBS: A Visual Basic script with embedded code.
  • MSI: Another type of Windows installer.
  • CMD: Similar to BAT files.
  • REG: Windows registry files.
  • WSF: A Windows Script File that permits mixed scripting languages.

You also need to keep an eye on Microsoft Office files with macros (such as DOCM, XLSM, and PPTM). Macros can be harmful but are also commonplace—especially in business documents. You’ll have to exercise your own judgment.

2. Encrypted Archive Files

As we just alluded to, archive files (such as ZIP, RAR, and 7Z) can conceal malware.

The problem is especially acute for encrypted archive files—i.e., those that require a password in order to extract their contents. Because they are encrypted, your email provider’s native antivirus scanner cannot see what they contain, and thus can’t flag it as malware.

The counterargument is that encrypted archive files are an excellent way to send sensitive data to a recipient; they are widely used for that purpose. Again, you’ll have to exercise your own judgment and make a decision about whether the file is safe.

3. Who Sent the Email?

It goes without saying that an email from a nonsensical address (for example, e34vcs@hotmail.com) is almost certainly something you shouldn’t open. Instead, immediately flag it as spam and remove it from your inbox.

That part is easy, but the situation can quickly become more complex.

Malicious actors are experts in making email addresses look like they are from an official source when in practice, they are phishing attacks. For instance, perhaps your bank’s email address is customers@bigbank.com; a hacker might send an email from customers@bigbank.co instead. That’s easy to overlook when you’re scanning through your inbox in a hurry.

There’s also been an uptick in email spoofing in recent years. When spoofing, an attacker tricks the email server into thinking the email came from the address being spoofed. You’ll even see the person’s real address and profile picture in the sender field.

In theory, you can spot spoofed emails by investigating the email’s source code, but it’s way beyond the abilities of most users. If you’re not expecting an email from the sender and the attached file ticks some of the other boxes we’re discussing, it’s probably malware.

Finally, remember that an attachment could be malicious even if you know the sender and the email is not spoofed. If the sender’s own machine is infected, it could send emails to their contact list without their knowledge.

4. Strange Filenames

In the same way that you should treat random email addresses with extreme distrust, so too should you be wary of attachments with filenames composed of random strings of characters.

People don’t save documents with a 20-character alphanumeric code as its name, and your computer would never prompt you to do so.

Similarly, names like “freemoney” or “greatopportunity” from an unknown sender are likely to contain malware and should immediately ring alarm bells.

5. Study the Contents of the Email

microsoft spam email

The text of the email can offer some clues about whether the message—and thus any attachment—is trustworthy.

Bots write many of the spam emails, spoofed emails, and phishing emails that you receive. They often have lousy formatting and spelling errors.

There are other little giveaways, too. For example, perhaps an email that’s purportedly from your best friend refers to you by your full name rather than your nickname. Or maybe it uses formal language and other syntax that you know the person in question would never use.

You should also be suspicious of an email that asks you to download and run its attachment. These emails are often made to appear as if they come from companies like FedEx and DHL; they claim that you can track your package via the download. Given that we live in an age where online shopping is routine, it’s easy to be duped, especially if you’re expecting deliveries.

6. Use Your Antivirus Suite

If you’re caught in two minds about the potential safety of an email attachment, make sure you always run it through your desktop antivirus app before running it on your machine.

Needless to say, if your antivirus program flags the file as suspicious, stop. Delete the file from your computer and don’t redownload it. The worst course of action would be to click through the various malware warnings and proceed regardless.

Remember, even though antivirus apps may not be perfect (they occasionally flag false positives), they are infinitely more trustworthy than a suspicious email which claims its attachment is safe even if it gets flagged by a scan.

(Note: We’ve explained how to test your antivirus app’s accuracy if you would like more information.)

Always Keep a Healthy Suspicion With Emails

Unfortunately, there’s not a one-size-fits-all solution for spotting unsafe email attachments. Broadly speaking, however, the higher the number of red flags the attachment ticks, the more likely it is to be a hazardous file.

If you’re unsure, reach out to the sender and ask for clarification. Most businesses and individuals will be only too happy to inform you about an attachment’s veracity or otherwise. Ultimately, stick to the golden rule: if in doubt, don’t proceed until you’re confident that it’s safe to do so.

If you’d like to learn more about staying safe while using email, take a few moments to learn how to stop spam email in Gmail and how to spot spear phishing email scams.

Read the full article: How to Spot Unsafe Email Attachments: 6 Red Flags


Read Full Article

Toolkit for digital abuse could help victims protect themselves


Domestic abuse comes in digital forms as well as physical and emotional, but a lack of tools to address this kind of behavior leaves many victims unprotected and desperate for help. This Cornell project aims to define and detect digital abuse in a systematic way.

Digital abuse may be many things: hacking the victim’s computer, using knowledge of passwords or personal date to impersonate them or interfere with their presence online, accessing photos to track their location, and so on. As with other forms of abuse, there are as many patterns as there are people who suffer from it.

But with something like emotional abuse, there are decades of studies and clinical approaches to address how to categorize and cope with it. Not so with newer phenomena like being hacked or stalked via social media. That means there’s little standard playbook for them, and both abused and those helping them are left scrambling for answers.

“Prior to this work, people were reporting that the abusers were very sophisticated hackers, and clients were receiving inconsistent advice. Some people were saying, ‘Throw your device out.’ Other people were saying, ‘Delete the app.’ But there wasn’t a clear understanding of how this abuse was happening and why it was happening,” explained Diana Freed, a doctoral student at Cornell Tech and co-author of a new paper about digital abuse.

“They were making their best efforts, but there was no uniform way to address this,” said co-author Sam Havron. “They were using Google to try to help clients with their abuse situations.”

Investigating this problem with the help of a National Science Foundation grant to examine the role of tech in domestic abuse, they and some professor collaborators at Cornell and NYU came up with a new approach.

There’s a standardized questionnaire to characterize the type of tech-based being experienced. It may not occur to someone who isn’t tech-savvy that their partner may know their passwords, or that there are social media settings they can use to prevent that partner from seeing their posts. This information and other data are added to a sort of digital presence diagram the team calls the “technograph” and which helps the victim visualize their technological assets and exposure.

technograph filled

The team also created a device they call the IPV Spyware Discovery, or ISDi. It’s basically spyware scanning software loaded on a device that can check the victim’s device without having to install anything. This is important because an abuser may have installed tracking software that would alert them if the victim is trying to remove it. Sound extreme? Not to people fighting a custody battle who can’t seem to escape the all-seeing eye of an abusive ex. And these spying tools are readily available for purchase.

“It’s consistent, it’s data-driven and it takes into account at each phase what the abuser will know if the client makes changes. This is giving people a more accurate way to make decisions and providing them with a comprehensive understanding of how things are happening,” explained Freed.

Even if the abuse can’t be instantly counteracted, it can be helpful simply to understand it and know that there are some steps that can be taken to help.

The authors have been piloting their work at New York’s Family Justice Centers, and following some testing have released the complete set of documents and tools for anyone to use.

This isn’t the team’s first piece of work on the topic — you can read their other papers and learn more about their ongoing research at the Intimate Partner Violence Tech Research program site.


Read Full Article

These robo-shorts are the precursor to a true soft exoskeleton


When someone says “robotic exoskeleton,” the power loaders from Aliens are what come to mind for most people (or at least me), but the real things will be much different: softer, smarter, and used for much more ordinary tasks. The latest such exo from Harvard is so low-profile you could wear it around the house.

Designed by researchers at Harvard’s Wyss Institute (in collaboration with several other institutions), which focuses on soft robotics and bio-inspired mechanisms, the exosuit isn’t for heavy lifting or combating xenomorphs but simply walking and running a little bit more easily.

The suit, which is really more of a pair of shorts with a mechanism attached at the lower back and cables going to straps on the legs, is intended to simply assist the leg in its hip-extension movement, common to most forms of locomotion.

An onboard computer (and neural network, naturally) detects the movements of the wearer’s body and determines both the type of gait (walking or running) and what phase of that gait the leg is currently in. It gives the leg making the movement a little boost, making it just that much easier to do it.

In testing, the suit reduced the metabolic load of walking by 9.3 percent and running by 4 percent. That might not sound like much, but they weren’t looking to create an Olympic-quality cyborg — just show reliable gains from a soft, portable exosuit.

“While the metabolic reductions we found are modest, our study demonstrates that it is possible to have a portable wearable robot assist more than just a single activity, helping to pave the way for these systems to become ubiquitous in our lives,” said lead study author Conor Walsh in a news release.

The whole idea, then, is to leave behind the idea of an exosuit as a big mechanical thing for heavy industry or work, and bring in the idea that one could help an elderly person stand up from a chair, or someone recovering from an accident walk farther without fatigue.

rt scitoc aug16 r1

The whole device, shorts and all, weighs about 5 kilograms, or 11 pounds. Most of that is in the little battery and motor pack stashed at the top of the shorts, near the body’s center of mass, helping it feel lighter than it is.

Of course this is the kind of thing the military is very interested in — not just for active duty (a soldier who can run twice as far or fast) but for treatment of the wounded. So it shouldn’t be a surprise that this came out of a DARPA project initiated years ago (and ongoing in other forms).

But by far the more promising applications are civilian, in the medical field and beyond. “We are excited to continue to apply it to a range of applications, including assisting those with gait impairments, industry workers at risk of injury performing physically strenuous tasks, or recreational weekend warriors,” said Walsh.

Currently the team is hard at work improving the robo-shorts, reducing the weight, making the assistance more powerful and more intuitive, and so on. The paper describing their system was the cover story of this week’s edition of the journal Science.


Read Full Article

The renaissance of silicon will create industry giants


Every time we binge on Netflix or install a new internet-connected doorbell to our home, we’re adding to a tidal wave of data. In just 10 years, bandwidth consumption has increased 100 fold, and it will only grow as we layer on the demands of artificial intelligence, virtual reality, robotics and self-driving cars. According to Intel, a single robo car will generate 4 terabytes of data in 90 minutes of driving. That’s more than 3 billion times the amount of data people use chatting, watching videos and engaging in other internet pastimes over a similar period.

Tech companies have responded by building massive data centers full of servers. But growth in data consumption is outpacing even the most ambitious infrastructure build outs. The bottom line: We’re not going to meet the increasing demand for data processing by relying on the same technology that got us here.

The key to data processing is, of course, semiconductors, the transistor-filled chips that power today’s computing industry. For the last several decades, engineers have been able to squeeze more and more transistors onto smaller and smaller silicon wafers — an Intel chip today now squeezes more than 1 billion transistors on a millimeter-sized piece of silicon.

This trend is commonly known as Moore’s Law, for the Intel co-founder Gordon Moore and his famous 1965 observation that the number of transistors on a chip doubles every year (later revised to every two years), thereby doubling the speed and capability of computers.

This exponential growth of power on ever-smaller chips has reliably driven our technology for the past 50 years or so. But Moore’s Law is coming to an end, due to an even more immutable law: material physics. It simply isn’t possible to squeeze more transistors onto the tiny silicon wafers that make up today’s processors.

Compounding matters, the general-purpose chip architecture in wide use today, known as x86, which has brought us to this point, isn’t optimized for computing applications that are now becoming popular.

That means we need a new computing architecture. Or, more likely, multiple new computer architectures. In fact, I predict that over the next few years we will see a flowering of new silicon architectures and designs that are built and optimized for specialized functions, including data intensity, the performance needs of artificial intelligence and machine learning and the low-power needs of so-called edge computing devices.

The new architects

We’re already seeing the roots of these newly specialized architectures on several fronts. These include Graphic Processing Units from Nvidia, Field Programmable Gate Arrays from Xilinx and Altera (acquired by Intel), smart network interface cards from Mellanox (acquired by Nvidia) and a new category of programmable processor called a Data Processing Unit (DPU) from Fungible, a startup Mayfield invested in.  DPUs are purpose-built to run all data-intensive workloads (networking, security, storage) and Fungible combines it with a full-stack platform for cloud data centers that works alongside the old workhorse CPU.

These and other purpose-designed silicon will become the engines for one or more workload-specific applications — everything from security to smart doorbells to driverless cars to data centers. And there will be new players in the market to drive these innovations and adoptions. In fact, over the next five years, I believe we’ll see entirely new semiconductor leaders emerge as these services grow and their performance becomes more critical.

Let’s start with the computing powerhouses of our increasingly connected age: data centers.

More and more, storage and computing are being done at the edge; that means, closer to where our devices need them. These include things like the facial recognition software in our doorbells or in-cloud gaming that’s rendered on our VR goggles. Edge computing allows these and other processes to happen within 10 milliseconds or less, which makes them more work for end users.

I commend the entrepreneurs who are putting the silicon back into Silicon Valley.

With the current arithmetic computations of x86 CPU architecture, deploying data services at scale, or at larger volumes, can be a challenge. Driverless cars need massive, data-center-level agility and speed. You don’t want a car buffering when a pedestrian is in the crosswalk. As our workload infrastructure — and the needs of things like driverless cars — becomes ever more data-centric (storing, retrieving and moving large data sets across machines), it requires a new kind of microprocessor.

Another area that requires new processing architectures is artificial intelligence, both in training AI and running inference (the process AI uses to infer things about data, like a smart doorbell recognizing the difference between an in-law and an intruder). Graphic Processing Units (GPUs), which were originally developed to handle gaming, have proven faster and more efficient at AI training and inference than traditional CPUs.

But in order to process AI workloads (both training and inference), for image classification, object detection, facial recognition and driverless cars, we will need specialized AI processors. The math needed to run these algorithms requires vector processing and floating-point computations at dramatically higher performance than general purpose CPUs provide.

Several startups are working on AI-specific chips, including SambaNova, Graphcore and Habana Labs. These companies have built new AI-specific chips for machine intelligence. They lower the cost of accelerating AI applications and dramatically increase performance. Conveniently, they also provide a software platform for use with their hardware. Of course, the big AI players like Google (with its custom Tensor Processing Unit chips) and Amazon (which has created an AI chip for its Echo smart speaker) are also creating their own architectures.

Finally, we have our proliferation of connected gadgets, also known as the Internet of Things (IoT). Many of our personal and home tools (such as thermostats, smoke detectors, toothbrushes and toasters) operate on ultra-low power.

The ARM processor, which is a family of CPUs, will be tasked for these roles. That’s because gadgets do not require computing complexity or a lot of power. The ARM architecture is perfectly designed for them. It’s made to handle smaller number of computing instructions, can operate at higher speeds (churning through many millions of instructions per second) and do it at a fraction of the power required for performing complex instructions. I even predict that ARM-based server microprocessors will finally become a reality in cloud data centers.

So with all the new work being done in silicon, we seem to be finally getting back to our original roots. I commend the entrepreneurs who are putting the silicon back into Silicon Valley. And I predict they will create new semiconductor giants.


Read Full Article

Corey Weiner is taking over as CEO of mobile ad company Jun Group


After 18 years at the helm, Mitchell Reichgut is stepping down as CEO of Jun Group, with COO and President Corey Weiner taking over as chief executive.

The news comes just about a year after Jun Group was acquired by Advantage Solutions, but Reichgut said the acquisition was a “non-factor” in his decision.

“I think it is the right time for the company to have a leadership change,” he said. “I have been stepping back more and more, so it’s a natural progression, with a bunch of managers here taking on larger roles as I move on.”

In addition to Weiner (who’s been at Jun Group since 2003), other Jun Group executives taking on new roles include Mishel Alon becoming COO, Leslie Bargmann becoming vice president of client services and Jeremy Ellison becoming vice president of technology.

Reichgut, meanwhile, said he’s “stepping back entirely to focus on artwork and writing and community service after a long, long career.”

Looking ahead, Weiner he plans to double down on Jun Group’s approach to advertising, where it builds custom audience segments by polling users in its network, then shows video ads and branded content to interested viewers.

“Our primary motivation is to evangelize that format,” he said. “As you know, most advertising is interruptive and consumers don’t like that kind of advertising very much — in some cases, they’re annoyed by it . This value exchange flips the advertising paradigm on its head. By choosing to engage with advertising, they are getting something amazing in return.”


Read Full Article