Improving On-Device Speech Recognition with VoiceFilter-Lite

Posted by Quan Wang, Software Engineer, Google Research

Voice assistive technologies, which enable users to employ voice commands to interact with their devices, rely on accurate speech recognition to ensure responsiveness to a specific user. But in many real-world use cases, the input to such technologies often consists of overlapping speech, which poses great challenges to many speech recognition algorithms. In 2018, we published a VoiceFilter system, which leverages Google’s Voice Match to personalize interaction with assistive technology by allowing people to enroll their voices.

While the VoiceFilter approach is highly successful, achieving a better source to distortion ratio (SDR) than conventional approaches, efficient on-device streaming speech recognition requires addressing restrictions such as model size, CPU and memory limitations, as well as battery usage considerations and latency minimization.

In “VoiceFilter-Lite: Streaming Targeted Voice Separation for On-Device Speech Recognition”, we present an update to VoiceFilter for on-device use that can significantly improve speech recognition in overlapping speech by leveraging the enrolled voice of a selected speaker. Importantly, this model can be easily integrated with existing on-device speech recognition applications, allowing the user to access voice assistive features under extremely noisy conditions even if an internet connection is unavailable. Our experiments show that a 2.2MB VoiceFilter-Lite model provides a 25.1% improvement to the word error rate (WER) on overlapping speech.

Improving On-Device Speech Recognition
While the original VoiceFilter system was very successful at separating a target speaker's speech signal from other overlapping sources, its model size, computational cost and latency are not feasible for speech recognition on mobile devices.

The new VoiceFilter-Lite system has been carefully designed to fit on-device applications. Instead of processing audio waveforms, VoiceFilter-Lite takes exactly the same input features as the speech recognition model (stacked log Mel-filterbanks), and directly enhances these features by filtering out components not belonging to the target speaker in real time. Together with several optimizations on network topologies, the number of runtime operations is drastically reduced. After quantizing the neural network with the TensorFlow Lite library, the model size is only 2.2 MB, which fits most on-device applications.

To train the VoiceFilter-Lite model, the filterbanks of the noisy speech are fed as input to the network together with an embedding vector that represents the identity of the target speaker (i.e., a d-vector). The network predicts a mask that is element-wise multiplied to the input to produce enhanced filterbanks. A loss function is defined to minimize the difference between the enhanced filterbanks and the filterbanks from the clean speech during training.

Model architecture of the VoiceFilter-Lite system.

VoiceFilter-Lite is a plug-and-play model, which allows the application in which it’s implemented to easily bypass it if the speaker did not enroll their voice. This also means that the speech recognition model and the VoiceFilter-Lite model can be separately trained and updated, which largely reduces engineering complexity in the deployment process.

As a plug-and-play model, VoiceFilter-Lite can be easily bypassed if the speaker did not enroll their voice.

Addressing the Challenge of Over-Suppression
When speech separation models are used for improving speech recognition, two types of error could occur: under-suppression, when the model fails to filter out noisy components from the signal; and over-suppression, when the model fails to preserve useful signal, resulting in some words being dropped from the recognized text. Over-suppression is especially problematic since modern speech recognition models are usually already trained with extensively augmented data (such as room simulation and SpecAugment), and thus are more robust to under-suppression.

VoiceFilter-Lite addresses the over-suppression issue with two novel approaches. First, it uses an asymmetric loss during the training process, such that the model is less tolerant to over-suppression than under-suppression. Second, it predicts the type of noise at runtime, and adaptively adjusts the suppression strength according to this prediction.

VoiceFilter-Lite adaptively applies stronger suppression strength when overlapping speech is detected.

With these two solutions, the VoiceFilter-Lite model retains great performance on streaming speech recognition for other scenarios, such as single-speaker speech under quiet or various noise conditions, while still providing significant improvement on overlapping speech. From our experiments, we observed a 25.1% improvement of word error rate after the 2.2MB VoiceFilter-Lite model is applied on additive overlapping speech. For reverberant overlapping speech, which is a more challenging task to simulate far-field devices such as smart home speakers, we also observed a 14.7% improvement of word error rate with VoiceFilter-Lite.

Future Work
While VoiceFilter-Lite has shown great promise for various on-device speech applications, we are also exploring several other directions to make VoiceFilter-Lite more useful. First, our current model is trained and evaluated with English speech only. We are excited about adopting the same technology to improve speech recognition for more languages. Second, we would like to directly optimize the speech recognition loss during the training of VoiceFilter-Lite, which can potentially further improve speech recognition beyond overlapping speech.

Acknowledgements
The research described in this post represents joint efforts from multiple teams within Google. Contributors include Quan Wang, Ignacio Lopez Moreno, Mert Saglam, Kevin Wilson, Alan Chiao, Renjie Liu, Yanzhang He, Wei Li, Jason Pelecanos, Philip Chao, Sinan Akay, John Han, Stephen Wu, Hannah Muckenhirn, Ye Jia, Zelin Wu, Yiteng Huang, Marily Nika, Jaclyn Konzelmann, Nino Tasca, and Alexander Gruenstein.

RIP Google Music, one of the company’s last examples of generosity

Google Music is dead, and with it one of the few remaining connections I have to the company that doesn’t feel like a gun to my head. The service, now merged haphazardly with YouTube Music, recalled the early days of Google, when they sometimes just made cool internet things. It made it nearly a decade, though — pretty impressive for a one of their products.

I’ll just say it up front: I’m a lifelong music pirate. Oh yes, I’ve reformed in recent years, but I’ve got a huge library of tracks that I’ve cultivated for decades and don’t plan to abandon any time soon (likewise you can pry Winamp from my cold, dead hands). So when Google announced back in 2011 I could stream it all to myself for free, it sounded too good to be true.

And indeed it was a relic of the old Google, which was quite simply all about taking things that are difficult to do yourself (find things online, set up a new email address, collaborate on a spreadsheet) and make them easier.

Google Music — as we’ll call it despite it having gone through several branding changes before the final indignity of being merged into another, worse service as a presumably short-lived tab — was not first to the music-streaming or downloading world by a long shot, but its promise of being able to upload your old music files and access them anywhere as if they were emails or documents was a surprisingly generous one.

Generous not just in that it was providing server space for 20,000 songs (!) for free and the infrastructure for serving those songs where you went, but in its acknowledgement of other models of owning media. It didn’t judge you for having 20,000 MP3s — they weren’t subjected to some kind of legitimacy check, and they didn’t report you to the RIAA for having them, though they certainly could have.

No, Google Music’s free media locker was the company, or at least a quorum of the product team, announcing that they get it: not everyone does everything the same way, and not everyone is ready to embrace whatever business model tech companies decide makes sense. (Notably it has shifted several times more since then.)

Though my perennial work frenemy at the time MG Siegler was not impressed with the beta, I vigorously defended it, noting that Google was starting simple and looking forward rather then trying to beat Apple at their own game. Plus, secretly, I was feverishly uploading a hundred gigs of music I’d gotten from Audiogalaxy, Napster, and SoulSeek. Here, I thought, was a bridge between my antiquarian habits and the cutting edge of tech.

Since then, like the resentfully loving owner of a junker, over the years I’ve been frustrated by Google Music in the ways that only one who truly relies on something can be. The app became essential to me even as its ever-changing and confusing interface confounded me. As Google’s media strategy and offerings fluctuated and blurred, my uploaded music sat there quietly, doing the same thing it did at launch: hosting my music files. Whatever it did in addition to that, it still let me access the glitchy, 128kbps version of The Bends I downloaded in 2001. I also had the security of knowing if my many drives died in a fire, I could at least recover my precious MP3s.

Whether I’d ripped it myself, pirated it in college, bought it on Bandcamp, or got it from a code inside the vinyl I bought at a show, it worked on Google Music. It integrated all my music in a truly all-accepting cloud player, and for that reason, I loved it in spite of its flaws and total lack of hipness.

Now, in deference to the explosion of YouTube’s popularity as a music platform — which more  than anything else really is due to a new type of laziness and platform agnosticism peculiar to the next generation — Google Music exists as a sort of ghost of itself within the YouTube Music app, itself an evolution of a couple other failed music strategies.

Perhaps Google felt that the optics of obsoleting a service and cutting off millions of users from something useful and beloved were not worth risking — after doing exactly that with Reader (RIP).

God Damn It, Google

So (after culling the users who forgot they had accounts) they settled on the next best thing, which was making Google Music suck. Buried inside the new app, the music I uploaded has undergone a regression: intermingled and mixed, poorly organized, unable to be searched through, and at every occasion presented as the worse option, the uploaded library function seems to have been hidden away and hobbled.

The ugly and reliable Music Manager, which has run in the background on my Windows PCs for years, is dead, and adding new music is done by manually dragging the files onto the YouTube Music tab. Complaining about having to move my fingers a few inches when I get a new album seems a bit pampered, so I’ll just say that it’s telling that Google chose to make the user do the work when the whole service was built around preventing exactly that kind of work from having to be done.

I suppose I’m an exception to the usual Google and YouTube user, and as I’ve been careful to show the company for the last 20 years or so, there’s no money to be made from me. Yet as soon as I understood that Google was going to make it hard for me to do what I had been doing with them for a decade, I decided I was willing to pay for it. Now I pay Plex for a service Google decided was below them, and incidentally it’s way better. (Come to think of it, I started paying for Feedly after Google killed Reader, too.)

In a way I’m thankful. The idea of divorcing myself entirely from Google’s ecosystem isn’t a realistic one for me, though I do it where I can (though having moved to iOS, the cure sometimes seems worse than the disease). One of the tattered bindings holding me to Google was the music thing. And while I do plan to take up a hundred gigabytes on one of their databases somewhere for as long as I possibly can, I’m glad the company admitted that what they were giving me didn’t make sense for them any more. It means one less reason that what Google has to give makes sense for me.

Every service from Google now, especially with those new, bad logos, feels less like it’s offering a solution to a problem and more like it’s just another form of leverage for the company. We were spoiled by the old, weird Google that did things like Books because they could, throwing it in the teeth of the publishers, or Wave, an experiment in interactivity that in many ways is still ahead of its time. They did things because they hadn’t been done, and now they do things because they can’t let you leave.

So, RIP Google Music. You were good while you lasted, but ultimately what you did best was show me that we deserved better, and we weren’t going to get it by waiting around for Google to return to its roots.

Read Full Article

Amazon’s new ‘Care Hub’ lets Alexa owners keep tabs on aging family members

Amazon today announced a set of new features aimed at making its Alexa devices more useful to aging adults. With the launch of “Care Hub,” an added option in the Alexa mobile app, family members can keep an eye on older parents and loved ones, with their permission, in order to receive general information about their activities and to be alerted if the loved one has called out for help.

The idea behind Care Hub, the company explains, is to offer reassurance to family members concerned about an elderly family member’s well-being, while also allowing those family members to maintain some independence.

This is not a novel use case for Alexa devices. Already, the devices are being used in senior living centers and other care facilities, by way of third-party providers.

K4Connect, a startup bringing tech to senior living centers, closes its $21M Series B

Amazon stresses that while family members will be able to keep an eye on their loved ones’ Alexa use, it will respect their privacy by not offering specific information. For example, while a family member may be able to see that their parent had played music, it won’t say which song was played. Instead, all activity is displayed by category.

In addition, users will be able to configure alerts if there’s no activity or when the first interaction with the device occurs on a daily basis.

And if the loved one calls for help, the family member designated as the emergency contact can drop in on them through the Care Hub or contact emergency services.

Image Credits: Amazon

These new features are double-opt in, meaning that both the family member and their loved one need to first establish a connection between their Alexa accounts through an invitation process. This is begun through the new Care Hub feature in the Alexa app, then confirmed via text message or email.

That may seem like a reasonable amount of privacy protection, but in reality, many older adults either struggle with or tend to avoid technology. Even things seemingly simple — like using a smartphone, email or texting — can sometimes be a challenge.

That means there are scenarios where a family member could set up the Care Hub system by accessing the other person’s accounts without their knowledge or by inventing an email that becomes “the parent’s email” just for this purpose.

Alternatively, they could just mislead mom or dad by saying they are helping them set up the new Alexa device, and — oh, can I borrow your phone to confirm something for the setup? (Or some other such deception.)

A more appropriate option to protect user privacy would be to have Alexa periodically ask the loved one if they were still okay with the Care Hub monitoring option being enabled, and to alert the loved one via the Alexa mobile app that a monitoring option was still turned on.

Of course, there may certainly be older adults who appreciate the ability to be connected to family in this way, especially if they are located at a distance from their family or are feeling isolated due to the coronavirus pandemic and social distancing requirements that’s keeping family members from being able to visit.

Amazon says Care Hub is rolling out in the U.S. The company notes it will learn from customer feedback to expand the feature set over time.

Read Full Article

10 Free Academic Databases for Students

Thanks to the internet and modern technology, doing research is much easier for modern students. Of course, the college library is great, but having access to peer-reviewed sources from all over the world on your laptop is on another level. Both undergraduate and graduate students benefit from academic sources online. The only problem is that […]

The post 10 Free Academic Databases for Students appeared first on ALL TECH BUZZ.

Come June 1, 2021, all of your new photos will count against your free Google storage

Come June 1, 2021, Google will change its storage policies for free accounts — and not for the better. Basically, if you’re on a free account and a semi-regular Google Photos user, get ready to pay up next year and subscribe to Google One.

Currently, every free Google Account comes with 15 GB of online storage for all your Gmail, Drive and Photos needs. Email and the files you store in Drive already counted against those 15 GB, but come June 1, all Docs, Sheets, Slides, Drawings, Forms or Jamboard files will count against the free storage as well. Those tend to be small files, but what’s maybe most important here, virtually all of your Photos uploads will now count against those 15 GB as well.

That’s a bid deal because today, Google Photos lets you store unlimited images (and unlimited video, if it’s in HD) for free as long as they are under 16MP in resolution or you opt to have Google degrade the quality. Come June of 2021, any new photo or video uploaded in high quality, which currently wouldn’t count against your allocation, will count against those free 15 GB.

Image Credits: Google

As people take more photos every year, that free allotment won’t last very long. Google argues that 80 percent of its users will have at least three years to reach those 15 GB. Given that you’re reading TechCrunch, though, chances are you’re in those 20 percent that will run out of space much faster (or you’re already on a Google One plan).

Some good news: to make this transition a bit easier, photos and videos uploaded in high quality before June 1, 2021 will not count toward the 15 GB of free storage. As usual, original quality images will continue to count against it, though. And if you own a Pixel device, even after June 1, you can still upload an unlimited number of high-quality images from those.

To let you see how long your current storage will last, Google will now show you personalized estimates, too, and come next June, the company will release a new free tool for Photos that lets you more easily manage your storage. It’ll also show you dark and blurry photos you may want to delete — but then, for a long time Google’s promise was you didn’t have to worry about storage (remember Google’s old Gmail motto? ‘Archive, don’t delete!’)

In addition to these storage updates, there’s a few additional changes worth knowing about. If your account is inactive in Gmail, Drive or Photos for more than two years, Google ‘may’ delete the content in that product. So if you use Gmail but don’t use Photos for two years because you use another service, Google may delete any old photos you had stored there. And if you stay over your storage limit for two years, Google “may delete your content across Gmail, Drive and Photos.”

Cutting back a free and (in some cases) unlimited service is never a great move. Google argues that it needs to make these changes to “continue to provide everyone with a great storage experience and to keep pace with the growing demand.”

People now upload more than 4.3 million GB to Gmail, Drive and Photos every day. That’s not cheap, I’m sure, but Google also controls every aspect of this and must have had some internal projections of how this would evolve when it first set those policies.

To some degree, though, this was maybe to be expected. This isn’t the freewheeling Google of 2010 anymore, after all. We’ve already seen some indications that Google may reserve some advanced features for Google One subscribers in Photos, for example. This new move will obviously push more people to pay for Google One and more money from Google One means a little bit less dependence on advertising for the company.

Read Full Article

Facebook extends its temporary ban on political ads for another month

The election is settled, but the nation is far from it.

Before Election Day in the U.S., Facebook hit pause on all political and social issue ads. At the time, the company made it clear that the precautionary measure designed to turn off one potential faucet of misinformation would be temporary, but it couldn’t say how long the policy would remain in effect.

Now, Facebook says the temporary ban will continue for at least another month. The decision to extend the special policy was implemented Wednesday, four days after Joe Biden’s election victory — and four days after it became clear that Trump had no intention of conceding a lost election.

“The temporary pause for ads about politics and social issues in the US continues to be in place as part of our ongoing efforts to protect the election,” the company wrote in an update to its previous announcement. “Advertisers can expect this to last another month, though there may be an opportunity to resume these ads sooner.”

Facebook’s ongoing political ad pause throws a wrench into things in Georgia, where two January runoff elections will decide which party will control the Senate heading into President-Elect Biden’s administration. A friendly Senate is essential for many of Biden’s biggest proposals, including a $2 trillion climate package that could reshape the American economy and push the country toward an electrified future that doesn’t rely on fossil fuels.

Over the last few days, a shocking number of Republicans have “humored” the president’s refusal to transfer power in spite of an unambiguous election call and Biden’s decisive win in Pennsylvania, which cut off any potential paths to victory for his opponent. The Trump campaign’s last-ditch flurry of legal challenges have presented little of substance so far, and they might ultimately be more about dividing a nation and sowing doubt than prevailing in court.

Close US election results plunge social media into nightmare misinformation scenario

Read Full Article

MTV partners with Unrd to create a mobile version of ‘Ghosted: Love Gone Missing’

Mobile storytelling startup Unrd is making its first move into adapting existing intellectual property — specifically “Ghosted: Love Gone Missing,” an MTV reality series about ghosting (the dating practice, not anything supernatural).

Until now, Unrd (pronounced “unread”) has created original crime, horror and romance stories that are told through characters’ phones, through content like text messages, video footage and more.

Starting next week, on November 16, the app will feature a version of “Ghosted” that — unlike the TV show — is scripted, as users explore characters’ text messages, photos and video calls to discover why they’ve been ghosted. They’ll even get to vote on whether the characters should “ghost” or “make up” before they see the stories’ ending (their votes won’t affect the outcome).

MTV Head of Digital Rory Brown told me that this was a “very close collaboration” between MTV and the Unrd team, led by CEO Shib Hussain.

“This is the first time they’ve partnered with an already established IP — but that didn’t scare us at all, to be that first media partner that they worked with,” Brown said. “There was a strong point of view on our side of the house how to keep it true to the existed format, while the Unrd team helped us reimagine it, and our collaboration met in the middle of that Venn diagram.”

Image Credits: Unrd

He also argued that while interactivity can be “a bit of a buzzword in the industry,” Unrd isn’t focusing on “interactivity for interactivity’s sake.” Instead, the aim is to create “a more immersive experience for the user.”

Unrd will feature three stories tied to “Ghosted,” each of them unfolding over six days.

“The key thing that we do different is this notion of real time,” Hussain said. “You can’t just binge it and consume every story in one day. You’ve got to wait with the character for the next message. That’s more immersive, and it also builds that tension and excitement amongst users as well.”

Brown noted that these Unrd stories are launching during a break in the second season of “Ghosted.” The hope is that they’ll keep existing fans engaged while creating new fans as well.

“At MTV, we’re always going to keep looking at ways to test the elasticity of IP,” he said. “I think Unrd is one way to do that. We’re talking to other partners, but Shib and his team have been fantastic to work with and we’d love to keep the relationship going.”

HBO releases a wellness-focused AR app to promote ‘His Dark Materials’

Read Full Article

India approves Google’s $4.5 billion deal with Reliance’s Jio Platforms

India’s antitrust watchdog has approved Google’s proposed investment of $4.5 billion in the nation’s largest telecom platform Jio Platforms, it said in a tweet on Wednesday.

Google announced in July that it would be investing $4.5 billion for a 7.73% stake in the top Indian telecom network. As part of the deal, Google and Jio Platforms plan to collaborate on developing a customized-version of Android mobile operating system to build low-cost, entry-level smartphones to serve the next hundreds of millions of users, the two companies said.

Jio Platforms is planning to launch as many as 200 million smartphones in the next three years, according to a pitch the telecom giant has made to several developers. These smartphones, as is the case with nearly 40 million of Jio’s feature phones in circulation today, will have an app store with only a few dozen apps, all vetted and approved by Jio, according to one developer who was pitched by Jio Platforms. An industry executive described Jio’s store as a walled garden.

Google to invest $10 billion in India

The Indian watchdog, Competition Commission of India (CCI), was said to be interested in reviewing the data sharing agreement between Google and Jio, Indian newspaper Economic Times reported last month, citing an unidentified source.

The announcement today comes days after the CCI announced it had directed an in-depth investigation into Google to verify the allegations of whether the Android-maker promotes its payments service during the installation of an Android smartphone (and whether phone vendors have a choice to avoid this); and if Google Play Store’s billing system is designed “to the disadvantage of both i.e. apps facilitating payment through UPI, as well as users.”

The call for this in-depth investigation was prompted after the CCI concluded in its initial review that requiring Google Pay to be used to buy apps or make in-app payments was an “imposition of unfair and discriminatory condition, denial of market access for competing apps of Google Pay and leveraging on the part of Google,” the watchdog said.

Jio Platforms, which has amassed over 400 million subscribers, has this year raised over $20 billion from 13 high-profile investors including Facebook, which alone invested $5.7 billion into the Indian firm. That deal has also been approved by the CCI. Jio Platforms is a subsidiary of Reliance Industries, India’s most valued firm. It is run by Mukesh Ambani, Asia’s richest man.

Read Full Article

How to Drive User Engagement to Your Site

User engagement on a website is a critical factor in ensuring that your website ranks well on search engine platforms. When my co-founders and I started Ascend Finance, I knew that user engagement could potentially be the most important metric to measure. Most websites have competitors that offer a similar product or service. Even if […]

The post How to Drive User Engagement to Your Site appeared first on ALL TECH BUZZ.

Josh.ai launches a ‘nearly invisible’ Amazon Echo competitor that’s the size of a coin

In the past several weeks we’ve seen refreshes and product expansions from about every facet of the smart home virtual assistant world. Apple launched the HomePod Mini, Google offered a long-overdue refresh of the Google Home and Amazon found even more speaker shapes to shove Alexa into.

Today, we’re getting an addition from a startup competitor. Josh.ai has aimed to build out a niche in the space by building a smart assistant product that’s designed to be professionally installed alongside other smart home wares, and they announced a new product this afternoon.

The device, Josh Nano, fully buys into a more luxury home-focused niche with a low-profile device that appears to be a little bit bigger than a half-dollar, though the bulk of the device is embedded into the wall itself and wired back to a central unit via power-over-ethernet. The device bundles a set of four microphones eschewing any onboard speaker, instead opting to integrate directly with a user’s at-home sound system. Josh boasts compatibility with most major AV receiver manufacturers in addition to partnerships with companies like Sonos. There isn’t much else to the device; a light for visual feedback, a multi-purpose touch sensor and a physical switch to cut power to the onboard microphones in case users want extra peace of mind.

Image via Josh.ai

The aim of the new hardware is to hide the smart features of a home and move away from industry-standard touchscreen hubs with dated interfaces. By stripping down a smart home product to its essential feature, Josh.ai hopes it can push more users to buy in more fully with confidence that subsequent hardware releases won’t render their devices outdated and ugly. The startup is taking pre-orders for the device (available in black and white color options) now and hopes to start shipping early next year.

Powering these devices is a product the company calls Josh Core, a small server which basically acts as a hub for everything Josh talks to in a user’s home, ensuring that interactions between smart home devices can occur locally, minimizing external requests. The startup will also continue selling its previously released Josh Micro, which integrates a dedicated speaker into the wall-mounted hardware.

Though Josh.ai partners directly with professional installers on the hardware, the startup has been scaling as a software business, offering consumers a license to their technology on an annual, five-year or lifetime basis. The price of that license also differs depending on what size home they are working with, with “small” rollouts being classified as homes with fewer than 15 rooms. In terms of hardware costs, Josh.ai says that pricing varies, but for most jobs, the average cost for users works out to be something like $500 per room.

Massive tech companies naturally design their products for massive audiences. For startups like Josh.ai this fact provides an in-road to design products that aren’t built for the common needs of a billion users. In fact, the selling point for plenty of their customers comes largely from the fact that they aren’t buying devices from Google, Amazon or Apple and hard-wiring microphones that feed back to them inside their home.

Though 95% of the startup’s business today focuses on residential, going forward, the company is also interested in scaling how their tech can be used in commercial scenarios like conference rooms or even elevators, the startup tells me.

Smart home startup Josh.ai raises $11 million to offer a home assistant alternative to Alexa

Read Full Article

Josh.ai launches a ‘nearly invisible’ Amazon Echo competitor that’s the size of a coin

In the past several weeks we’ve seen refreshes and product expansions from about every facet of the smart home virtual assistant world. Apple launched the HomePod Mini, Google offered a long-overdue refresh of the Google Home, and Amazon found even more speaker shapes to shove Alexa into.

Today, we’re getting an addition from a startup competitor. Josh.ai has aimed to build out a niche in the space by building a smart assistant product that’s designed to be professionally installed alongside other smart home wares and they announced a new product this afternoon.

The device, Josh Nano, fully buys into a more luxury home-focused niche with a low-profile device that appears to be a little bit bigger than a half-dollar, though the bulk of the device is embedded into the wall itself and wired back to a central unit via power-over-ethernet. The device bundles a set of four microphones eschewing any onboard speaker, instead opting to integrate directly with a user’s at-home sound system. Josh boasts compatibility with most major AV receiver manufacturers in addition to partnerships with companies like Sonos. There isn’t much else to the device, a light for visual feedback, a multi-purpose touch sensor, and a physical switch to cut power to the onboard microphones in case users want extra peace of mind.

Image via Josh.ai

The aim of the new hardware is to hide the smart features of a home and move away from industry standard touch screen hubs with dated interfaces. By stripping down a smart home product to its essential feature, Josh.ai hopes it can push more users to buy in more fully with confidence that subsequent hardware releases won’t render their devices outdated and ugly. The startup is taking pre-orders for the device (available in black and white color options) now and hopes to start shipping early next year.

Powering these devices is a product the company calls Josh Core, a small server which basically acts as a hub for everything Josh talks to in a user’s home, ensuring that interactions between smart home devices can occur locally, minimizing external requests. The startup will also continue selling its previously released Josh Micro which integrates a dedicated speaker into the wall-mounted hardware.

Though Josh.ai partners directly with professional installers on the hardware, the startup has been scaling as a software business, offering consumers a license to their technology on an annual, 5-year or lifetime basis. The price of that license also differs depending on what size home they are working with, with “small” rollouts being classified as homes with fewer than 15 rooms. In terms of hardware costs, Josh.ai says that pricing varies but for most jobs, the average cost for users works out to be something like $500 per room.

Massive tech companies naturally design their products for massive audiences. For startups like Josh.ai this fact provides an in-road to design products that aren’t built for the common needs of a billion users. In fact, the selling point for plenty of their customers comes largely from the fact that they aren’t buying devices from Google, Amazon or Apple and hard-wiring microphones that feed back to them inside their home.

Though 95% of the startup’s business today focuses on residential, going forward, the company is also interested in scaling how their tech can be used in commercial scenarios like conference rooms or even elevators, the startup tells me.

Smart home startup Josh.ai raises $11 million to offer a home assistant alternative to Alexa

Read Full Article

Google adds COVID-related health and safety info to Google Travel

Google today announced it’s adding more COVID-related health and safety information to its Google Travel booking service. Starting this week, when users search for hotels and vacation rental properties through Google Travel, they may see new information about COVID-19 safety precautions at the property — like enhanced cleaning procedures that may be in use, for example, or if there’s an option for a contact-free check-in, among other things.

The company said it’s working in partnership with hotel chains, vacation rental providers and other industry associations to source this health and safety data.

Individual hotel owners will be able to directly provide Google with their COVID-19 health and safety procedures via a new tool in Google My Business that lets them add health and safety information to their Business Profile.

At launch, several leading hotel chains are also partnering with Google on the new effort, including Marriott, Hilton, Comfort Inn (Choice Hotels), Holiday Inn (IHG), Best Western, Hyatt, Radisson and Four Seasons, the company tells TechCrunch.

Image Credits: Google Travel, screenshot via TechCrunch

On the vacation rental side, Google is working with a range of partners, including large vacation rental providers, medium-sized distributors and online travel agents, who are able to provide this health and safety information directly.

VRBO is among those partners, but Google says it’s not working with individual property owners at this time. It’s also not sourcing any of this data from entries in Google My Business, as it is with hotels — only what’s directly shared by partners.

Currently, this new COVID-related health and safety information is only available when users search for accommodations on google.com/travel — and not on other Google services, like Google Maps.

This update follows others Google made earlier this year in response to the coronavirus pandemic. Today, the company allows users to find hotels and vacation rentals with refundable rates on Google Travel, as well as track hotel and flight availability in a given area.

In September, Google also added a COVID-19 layer to Google Maps, but this provides data about the number of cases per 100,000 across a given map area. Earlier, it had added COVID-19 checkpoints and other transit data to Maps, as well as features to help businesses survive government shutdowns.

The health and safety information will roll out to Google Travel users worldwide.

Google Maps gets a COVID-19 layer

 

Read Full Article

The Boox Poke 3 is my new favorite e-reader

There are plenty of e-readers to choose from out there, but never enough for me. I’m always questing for the one that will make me forget that there are others available, and in the Onyx Boox Poke 3, I think I have found it — at least for now. The Chinese e-paper device maker has nailed the size, the screen, and added a sprinkle of versatility that I didn’t know I was lacking.

The Poke 3 fits in the same “original flavor e-reader” category as the Kindle Paperwhite and Kobo Clara HD: 6 inches, 300 PPI or so (which makes for very clear text), and somewhere between $100 and $200.

These readers fit easily in a pocket, unlike the larger Oasis and even larger Forma; they tend to lack anything but a power button and are very focused on books and saved articles.

Kobo’s new entry-level Clara HD e-reader has a crisp, color-adjustable display

But the Kindle and Clara both have major flaws. The Kindle is tied to Amazon in all the ways I can’t stand, including on-device ads by default, and the Clara… well, beside the screen, the hardware is honestly just bad. Kobo made my previous favorite e-reading device, the compact and flush-front Aura, and I’ve finally found a worthy successor to that beloved gadget.

The Poke 3 is the latest device to come from Boox, the e-reader line from parent company Onyx. The company has mostly been a presence in Southeast Asia, particularly its home country of China, so don’t be surprised if you’ve never heard of it. Boox makes a wide variety of e-paper devices (which I will evaluate in a separate article), and the Poke 3 is the simplest and smallest of them.

Image Credits: Devin Coldewey / TechCrunch

I’ll highlight the device’s strengths first. Most importantly, it is a lovely piece of hardware. The flush front is a pleasure to read on, as there is no raised bezel to shade the text or collect grime. The power button is well located and clicky. There’s just enough border to hold onto without worrying about smudging or activating the screen, and a bit of extra space at the bottom enables plenty of comfortable grips.

It’s thinner than the competition and the build quality is excellent. The front is hardened glass from partner Asahi, which will hopefully prevent it from needing a cover.

Image Credits: Devin Coldewey / TechCrunch

The finish is, to be honest, something of a fingerprint and oil magnet, and could be grippier. The Paperwhite has it beat on texture but I prefer the smooth back to the weird perforated one of the Clara.

At 150 grams it’s 16 lighter than the Clara and 32 lighter than the Paperwhite. That doesn’t sound like a lot, but when you’re holding a device for hours straight every little bit counts, and at this size it also helps with balance.

Its 6-inch screen is not meaningfully different from the Kindle or Kobo devices out there in terms of resolution or font rendering. I scrutinized the Poke 3 next to the Clara HD and Forma and found no differences that anyone would notice reading from 10-20 inches away.

Image Credits: Devin Coldewey / TechCrunch

It does differ in its approach to illumination, though whether meaningfully so is a matter of opinion. Instead of having a brightness slider and a temperature slider, it has a warm and cool slider, and increasing or decreasing either one changes both the brightness and temperature. You can also turn either one off entirely or link them together so they are adjusted as one.

If it sounds more complicated… it is. I don’t see that it adds any real new capabilities, but once you get the feel for it, it isn’t that much harder to use, either. I do wish that when you linked the two sliders, they kept their positions relative to one another. The whole system seems a little baroque and I hope Boox streamlines it. That said, the quality of the light is equally good and once you dial it in, it looks great.

Type formatting is good, and has plenty of options for tweaking how any of the many (too many…) included fonts look, even weight and contrast adjustments to really fine tune them. Adding custom fonts is as easy as dragging and dropping them, just like documents.

The operating system of the Boox provides far more options than either Kobo or Kindle. Amazon keeps tight control over its ecosystem and outside of a handful of associated services the devices can’t do much. Kobo at least allows for more file formats to be loaded directly on, and now has excellent Pocket integration for saving articles from the web. Boox takes things two steps further with a custom Android launcher that you can download full apps onto.

Image Credits: Devin Coldewey / TechCrunch

Now, there are really only so many apps that you actually might want on an e-reader like this one. And not everything works as well as I’d like. But for the first time I can actually get Simplenote on my e-reader.

It’s not as simple as it would be on an ordinary Android device, though. Because the Poke 3 comes from China, it doesn’t have access to Google services right off the bat. You can add it through settings, which isn’t hard, but there’s also a sideloading store built in with recent (if not quite brand new) install packages of popular, vetted apps for the device.

Let’s just admit right now that compared to the simplicity of Kindle and Kobo, this is already a bit out there. And whether you feel comfortable logging into a version of Evernote that you can’t (without a bit of work) verify the contents of… well, it’s not for everyone. But to be clear on this, Boox isn’t some fly-by-night operation — they may not be well known over here, but it’s hard to argue with the quality of the devices. The problem is simply that localizing an OS built for users in China has some fundamental challenges.

Image Credits: Devin Coldewey / TechCrunch

Fortunately for everyone, the basic capability to load books on there and read them is solid and it’s what you’d be doing most of the time. There may be a busy interface when you’re doing other stuff, but you can easily hide all indicators like progress and title while you’re reading, dedicating every square inch of the screen to actual reading.

It has 32 gigs of internal memory, making storage of audiobooks (it has Bluetooth for sound) and bulky documents easy, and connects quickly as a drive when you plug in its USB-C cord.

The Poke 3 will cost $189 when it ships next week, which is on the high end for this type of device. That’s $30 more than a Kindle Paperwhite and $70 more than a Clara HD. But I honestly think it is worth the premium. This is a better e-reader, period; despite the sometimes fussy interface, I enjoy using it, and appreciate that it provides capabilities that its competition doesn’t. If you need simple and don’t mind a cheap build, the Clara is a great cheaper option, but for a step up consider the Poke 3.

The reMarkable 2 improves on the original in every way, but remains firmly in its niche

Read Full Article

Why I left edtech and got into gaming

Darshan Somashekar Contributor

Darshan Somashekar co-founded drop.io (acquired by Facebook), Imagine Easy Solutions (acquired by Chegg) and is currently co-founder of gaming platform Solitaired and a venture partner at TMV.

Now that COVID-19 has accelerated the adoption of digital education tools, edtech has become one of the hottest areas of investment.

As someone who has been in edtech for nearly 20 years, this sounds like the precise moment to capitalize on all the newfound interest. Which is why what I’m about to say might be surprising: I’m leaving edtech for the world of gaming with my new company, Solitaired.

I first got into edtech in high school, when a friend and I founded EasyBib, a website that helped students cite sources for their papers. At the time, we were just students who felt there had to be a better way than formatting tedious citations for research papers by hand. But as we dove into the business further, we realized there was a lot to like about bibliographies and education technology in general.

For one, the education market is large. There are more than 56 million K-16 students in the U.S., and over 1.3 billion globally. Federal, state and local governments spend an aggregate of 5% of GDP on education, and that doesn’t even include what students and parents spend on content and technology.

Secondly, it’s structured. Students generally all go through the same curriculum together. That means most students have the same problem in the same way; if you solve a problem for one group of users, you’ve probably solved it for most users.

The citation problem was just like that. When we sold our company to Chegg, we were already reaching four out of five students that needed bibliographies, or over 30 million students in the U.S. Edtech companies that help students with math, chemistry, homework help, tutoring and other curricular needs can build massive audiences quickly.

Edtech that’s part of the curriculum also has high engagement. EasyBib users stayed on our site for nearly ten minutes per session, creating one citation after another for their bibliographies. For direct-to-consumer edtech companies that are ad and subscription driven, this behavior creates many monetization opportunities.

While we grew fast, our endemic market opportunity was limited. Why? The strengths of edtech can also be its downsides, especially for a startup. On the user growth front, we focused on school relationships, marketing and SEO. But once we reached four out of every five students in the U.S., there wasn’t much more room to grow.

To increase engagement even further, we tried a number of things: encouraging more citation creation, adding research and note-taking features and building a Chrome extension to be more ever-present in the user’s research journey. Those efforts fell short too. Ultimately, the school calendar dictated how often students needed to use us, and we were constrained by the number of research papers teachers assigned.

These challenges can certainly be overcome. But as a startup, we had to decide if we wanted to pursue adjacencies and expansions ourselves. Ultimately, this realization was one of the reasons we decided to sell our company to Chegg, which had a wider user base and product synergies that we couldn’t achieve on our own. As anyone who follows Chegg might know, they’ve been very successful in accelerating the edtech digital transformation.

When we began thinking about our second business, we had these lessons in the back of our mind. That’s when we discovered gaming.

Read Full Article

Apple updates Mac Mini with Apple-designed M1 chip

In addition to the MacBook Air, Apple also introduced a new Mac Mini. It runs Apple’s first custom chipset specifically designed for Mac computers, the M1. It still features the same familiar design, but it now uses an Arm-based chip.

“If you compare a Mac mini to the top-selling desktop in this price range, the Mini is one-tenth its size and yet it’s up to five times faster,” Hardware Engineering Senior Manager Julie Broms said.

The M1 features four high-performance cores and four high-efficiency cores. It uses a 5nm process. There are eight GPU cores with 128 execution units, which should lead to peak performance of 2.6 teraflops. There is also Apple’s neural engine for machine learning tasks.

According to Apple, the new Mac Mini is up to three times faster than the previous model when it comes to CPU performance. As for GPU performance, it is up to six times faster. ML acceleration is also up to 15 times faster.

At the back of the device, you’ll find an Ethernet port, two Thunderbolt/USB4 ports that support a 6K display, an HDMI 2.0 port, two USB A ports and a 3.5mm headphone port.

Pricing starts at $699 and the new Mac Mini will be available next week.

Image Credits: Apple

Apple announces the M1, the first chip in its Apple Silicon family

This is the new $999 MacBook Air, powered by Apple silicon

Read Full Article

Apple will release macOS Big Sur on November 12

Apple’s upcoming desktop and laptop operating system, macOS Big Sur, will be released on November 12, the company announced today.

macOS Big Sur — which stays with the company’s California-themed naming scheme — will arrive with a new and refreshed user interface, new features, and performance improvements.

Much of the features in iOS 14 are porting over — including improved Message threading and in-line replies and a redesigned Maps app. The new Apple software also comes with a new Control Center, with quick access to brightness, volume, Wi-Fi, and Bluetooth.

Safari also gets a much needed lick of paint. It comes with new privacy and security features, including an in-built intelligence tracking prevention that stops trackers following you across the web, and password monitoring to save you from using previously breached passwords.

If you’re wondering what macOS Big Sir is like to work on, TechCrunch’s Brian Heater took the new software for a spin in August.

macOS Big Sur will be supported on Macs and MacBooks dating back to 2013.

Read more:

• Apple is building its own processors for future Macs
• What we expect from Apple’s ‘One More Thing’ Mac event
• macOS 11.0 Big Sur preview
• More from Apple’s November Mac Event

macOS 11.0 Big Sur preview

Read Full Article

Apple will release macOS Big Sur on November 12

Apple’s upcoming desktop and laptop operating system, macOS Big Sur, will be released on November 12, the company announced today.

macOS Big Sur — which stays with the company’s California-themed naming scheme — will arrive with a new and refreshed user interface, new features, and performance improvements.

Much of the features in iOS 14 are porting over — including improved Message threading and in-line replies and a redesigned Maps app. The new Apple software also comes with a new Control Center, with quick access to brightness, volume, Wi-Fi, and Bluetooth.

Safari also gets a much needed lick of paint. It comes with new privacy and security features, including an in-built intelligence tracking prevention that stops trackers following you across the web, and password monitoring to save you from using previously breached passwords.

If you’re wondering what macOS Big Sir is like to work on, TechCrunch’s Brian Heater took the new software for a spin in August.

macOS Big Sur will be supported on Macs and MacBooks dating back to 2013.

Read more:

• Apple is building its own processors for future Macs
• What we expect from Apple’s ‘One More Thing’ Mac event
• macOS 11.0 Big Sur preview
• More from Apple’s November Mac Event

macOS 11.0 Big Sur preview

Read Full Article

Huawei reportedly set to sell Honor budget phone division for $15B

Following weeks of rumors surrounding a potential sale, Huawei has reportedly struck a deal to divest itself of its Honor brand. A new report out today from Reuters notes that the embattled hardware maker will sell the budget unit to a consortium of buyers that includes the government of the city of Shenzhen and Digital China, a phone distributor.

The report, which cites “people familiar with the matter,” puts the price tag for the Honor unit at $15.2 billion. Honor’s new owners will reportedly keep much of the brand’s 7,000 employees (management included) in tact, with plans to take the company public in around three years. The Honor brand has largely focused on low-cost devices, with sales in China, Europe and the U.S. This deal would likely find Huawei focusing more exclusively on high-end products under its own brand.

While the deal has been rumored for some time now, its seeming conclusion comes in the wake of Joe Biden’s presidential election win. It seems clear from Huawei’s decision to press on with the all-cash deal that the company doesn’t believe its international fortunes will change immediately under a new U.S. president.

The news comes in the wake of ongoing difficulties tied to U.S. sanctions. Huawei’s inability to access technologies from companies like Google have proven to be a major hit for the world’s second largest phone maker. While the company has managed to maintain solid sales in its native China, even those have taken a hit amid its struggles.

Huawei plans to divest budget phone unit Honor: reports

Read Full Article