Baidu’s research lab announces “Deep Speech” recognition system

[TaiShang]

Baidu’s research lab announces “Deep Speech” recognition system
3:15 pm on Dec 19, 2014

Baidu Research, the research division of search giant Baidu, unveiled last night a speech recognition technology it has dubbed “Deep Speech.”

According to an
official release, Baidu’s team of artificial intelligence researchers in Sunnyvale designed Deep Speech to recognize and interpret voice input in noisy environments like restaurants, where ambient noise or other factors can muddle accuracy.

“Deep learning, trained on a huge dataset — over 100,000 hours of synthesized data — is letting us achieve significant improvements in speech recognition,” said Andrew Ng, chief scientist at Baidu Research, in a statement. “I’m excited by this progress, because I believe speech will transform mobile devices, as well as the Internet of Things. This is just the beginning.”

For years, Baidu has been researching deep learning and artificial intelligence as it continues to build out its search technology and mobile product suite in China. Earlier this may, the company announced it will invest US$300 million in its Sunnyvale facility, for which the company poached Coursera co-founder and Stanford professor Andrew Ng to lead.

Beyond Deep Search, Baidu has unveiled a number of projects – some public, others internal – that harness artificial intelligence. Speaking to the Wall Street Journal last November, Ng described Baidu Eye and the company’s partially-automated cars as “research explorations” rather than products. Earlier this year, the company also released an image search feature in its translation app that (with mixed results) identified objects in photographs.

 

[indiatester]

Is this open sourced? Can I adapt it to Indian languages?

 

[TaiShang]

Baidu Research Achieves Speech Recognition Breakthrough With "Deep Speech"

New Scalable Deep Learning System Aims to Improve Speech Recognition in Noisy Environments

Baidu December 18, 2014 9:00 AM

Dr. Andrew Ng, Chief Scientist, Baidu Click here for high-resolution version

SUNNYVALE, CA--(Marketwired - Dec 18, 2014) -Baidu Research, a division of Baidu, Inc. (NASDAQ:BIDU), today announced initial results from its Deep Speech speech recognition system.


Deep Speech is a new system for speech, built with the goal of improving accuracy in noisy environments (for example, restaurants, cars and public transportation), as well as other challenging environments (highly reverberant and far-field situations).

Key to the Deep Speech approach is a well-optimized recurrent neural net (RNN) training system that uses multiple GPUs, as well as a set of novel data synthesis techniques that allowed Baidu researchers to efficiently obtain a large amount of varied data for training.

Earlier this month, tests demonstrated the following:

• Deep Speech outperformed previously published results on the widely studied Switchboard Hub5'00 benchmark, obtaining 16.5% Word Error Rate.
• Deep Speech outperformed public web APIs (Google Web Speech, wit.ai) as well as commercial systems (Bing Speech Services, Apple Dictation), especially in the regime of speech against noisy backgrounds. In noisy environments, Deep Speech outperforms all of these systems by over 10% (Word Error Rate).

Dr. Andrew Ng, Chief Scientist at Baidu, commented: "Deep learning, trained on a huge dataset -- over 100,000 hours of synthesized data -- is letting us achieve significant improvements in speech recognition. I'm excited by this progress, because I believe speech will transform mobile devices, as well as the Internet of Things. This is just the beginning."

Dr. Dan Jurafsky, Professor of Linguistics and Computer Science at Stanford University, said: "I am enthusiastic about Baidu's new methods for speech recognition, especially the use of elegant models that make the problem simpler and easier to engineer, together with GPUs for speed and scalability. The results suggest some really exciting near-term directions for speech recognition, especially for noisy environments and challenging speech tasks."

"Speech recognition in noisy and reverberant conditions is still a challenging task for state-of-the-art speech recognition systems. This recent work by Baidu Research has the potential to disrupt how speech recognition will be performed in the future," added Dr. Ian Lane, Assistant Research Professor of Engineering, Carnegie Mellon University. "Baidu's innovative work with GPU scaling and large data sets brings us a step closer to the vision of being able to converse naturally with smart devices, appliances, wearables and robots, even in noisy environments."

The Deep Speech results are posted in Deep Speech: Scaling Up End-to-End Speech Recognition.

About Baidu Research: Baidu Research, based in Silicon Valley and Beijing, is led by Dr. Andrew Ng, Chief Scientist. Baidu Research comprises three interrelated labs: the Silicon Valley AI Lab, the Institute of Deep Learning and the Big Data Lab, led by Dr. Adam Coates, Dr. Kai Yu and Dr. Tong Zhang, respectively. The organization brings together global research talent to work on fundamental technologies in areas such as image recognition and image-based search, voice recognition, natural language processing and semantic intelligence (http://research.baidu.com).

Is this open sourced? Can I adapt it to Indian languages?

Cannot find info on this. It is definitely to be commercialized. But I am not familiar much with the details.

 

[Evil Flare]

meanwhile BAIDU storage is blocked in Pakistan ....

 

[TaiShang]

Baidu Institute of Deep Learning.

Baidu Institute of Deep Learning

Chinese Search Engine Baidu Confirms Strategic Investment In Uber, Will Add Ride Requests
Posted Dec 16, 2014

It’s official: Chinese Internet giant Baidu has confirmed that it is Uber’s latest investor, asTechCrunch reported last week. Baidu’s mobile search and maps apps will also integrate Uber “request a ride” buttons.

The companies did not disclose the amount of the investment. Bloomberg previously reported that Baidu had taken the entire $600 million surplus that Uber built into its most recent funding round, however a source at Baidu told TechCrunch that this is untrue.

If there is further capacity in the round, which was earmarked towards developing Uber’s market position across Asia Pacific, then it stands to reason that other strategic investors from Asia may make similar deals with the U.S. company.

We understand that the strategic partnership between Uber and Baidu is more significant than the investment, because Baidu Maps will be able to integrate Uber. This is similar toGoogle Maps’ Uber integration after Google Ventures invested in the car-calling app. Furthermore, Baidu’s mobile search app will be configured so that Uber is displayed prominently when users make travel- or venue-related queries.

The deal will “enable users of Baidu Map and Mobile Baidu, Baidu’s flagship mobile search app, to connect easily with Uber driver-partners” the companies said.

This is a major boon for Uber because Baidu operates China’s largest search engine and will help it compete against its rivals Didi Dache and rival Kuaidi Dache, which are backed by Baidu-competitors Tencent and Alibaba, respectively.

The Baidu investment comes just a few weeks after Uber announced that it had raised $1.2 billion in funding at a $40 billion valuation, with a portion of the capital earmarked for expansion in the Asia-Pacific region.

The deal was struck at a “signing ceremony” at Baidu HQ in Beijing, which Baidu chairman and CEO Robin Li and Uber CEO Travis Kalanick both attended.




The deal will also allow Uber to take advantage of Baidu’s app distribution channels, which include 91 Wireless, which it purchased for $1.9 billion last year. This is important because Google Play isn’t available in China and 91 Wireless runs some of the largest alternative app stores in the country. Baidu claims it is currently China’s largest mobile app distribution platform, and distributed an average of 160 million apps per day.

The company’s mobile search products currently have over 500 million monthly active users, while Baidu Map has over 240 million monthly active users, which has allowed it tocreate heat maps of travel patterns by users, similar to Uber’s God View.

Baidu is also currently tackling an expansion into Latin America via Brazil, which overlaps with Uber’s global aspirations.

In a prepared statement, Kalanick said “This collaboration marks a milestone for Uber. We’re currently in 250 cities around the world, and the Asia-Pacific region has been a key area of growth for us. Our partnership with Baidu—a premier global brand—reflects our commitment to the region and the growing community of Uber riders and driver-partners here.”

Uber is currently available in nine Chinese cities, including Beijing, Shanghai, Tianjin, Chongqing, Shenzhen, Guangzhou, Wuhan, Chengdu, and Hangzhou.

For more information about how Baidu and Uber can work together, see our previous article.

 

[TaiShang]

Baidu now has an eye in the sky with launch of new flight tracker app

There are 904 airplanes in Chinese airspace right now.

The app, called Baidu Tianyan (tianyan could be translated as “eye in the sky”), will be familiar to anyone who uses FlightAware, the app that was in the news earlier this year as people sought to understand what on earth happened to the still missing Malaysia Airlines flight MH370. As with FlightAware, Baidu’s new app is accompanied by a browser-based flight tracker – see
here – that does away with some of the slickness of the mobile app.

While the augmented reality aspect of the Baidu Tianyan mobile app is fun (pictured top), it’s actually the least useful part. It’s probably more useful and insightful when used to track flights on a map.:

You can also search by flight numbers to see if your flight is delayed. Or you could just marvel about how many sleek hunks of metal are screaming through the skies at the moment:

While it looks good and could prove useful, this only covers planes in or near Chinese airspaces, so this isn’t a global app yet.

Kaiser Kuo, Baidu’s director of international communications, explains that the new app tracks flights by their ADS-B signal (that’s the automatic dependent surveillance-broadcast system) and even tracks private aircraft.

 

[Lux de Veritas]

Andrew Ng is born in Hong Kong and he spent his formative years in Singapore, from Primary school to high school in our most prestigious institution.

Today he is No 1 expert in AI.

 

[bolo]

Andrew Ng is born in Hong Kong and he spent his formative years in Singapore, from Primary school to high school in our most prestigious institution.

Today he is No 1 expert in AI.

LOL, a singaporean taking credit for someone's achievement because he went to primary and junior high here? You went to school there, what happened to you?

 

[opruh]

Amazing

 

[Reem]

Is this open sourced? Can I adapt it to Indian languages?

Nope I don't think Baidu Deep Speech open source or developer API is released as yet. Although article below shows that it already adequately performs for English infact higher accuracy then other commercial speech apis from (Google, Bing, Apple) in noise prone situations.
http://nextbigfuture.com/2014/12/baidu-deep-speech-system-81-accurate-in.html

 

[Lure]

What Andrew Ng working for Baidu? This guy's video lectures was the thing that made me love pattern recognition and was very informative. This guy is one of the most brilliant academicians on the field. Clearly he will give a lot of value to Baidu.

 

[TaiShang]

What Andrew Ng working for Baidu? This guy's video lectures was the thing that made me love pattern recognition and was very informative. This guy is one of the most brilliant academicians on the field. Clearly he will give a lot of value to Baidu.

I guess this below interview gives an idea

Baidu’s Andrew Ng on Deep Learning and Innovation in Silicon Valley
Nov 21, 2014

• Six months ago, Chinese Internet-search giant Baidu signaled its ambitions to innovate by opening an artificial-intelligence center in Silicon Valley, in Google GOOGL -0.34%’s backyard. To drive home the point, Baidu hired Stanford researcher Andrew Ng, the founder of Google’s artificial-intelligence effort, to head it.

• Ng is a leading voice in “deep learning,” a branch of artificial intelligence in which scientists try to get computers to “learn” for themselves by processing massive amounts of data. He was part of a team that in 2012 famously taught a network of computers to recognize cats after being shown millions of photos.
  
  On a practical level, the field helps computers better recognize spoken words, text and shapes, providing users with better Web searches, suggested photo tags or communication with virtual assistants like Apple AAPL -0.47%’s Siri.
  
  In an interview with The Wall Street Journal, Ng discussed his team’s progress, the quirks of Chinese Web-search queries, the challenges of driverless cars and what it’s like to work for Baidu. Edited excerpts follow:
  
  WSJ: In May, we wrote about Baidu’s plans to invest $300 million in this facility and hire almost 200 employees. How’s that coming along?
  
  Ng: We’re on track to close out the year with 96 people in this office, employees plus contractors. We’re still doing the 2015 planning, but I think we’ll quite likely double again in 2015. We’re creating models much faster than I have before so that’s been really nice. Our machine-learning team has been developing a few ideas, looking a lot at speech recognition, also looking a bit at computer vision.
  
  WSJ: Are there examples of the team’s work on speech recognition and computer vision?
  
  Ng: Baidu’s performance at speech recognition has already improved substantially in the past year because of deep learning. About 10% of our web search queries today come in through voice search. Large parts of China are still a developing economy. If you’re illiterate, you can’t type, so enabling users to speak to us is critical for helping them find information. In China, some users are less sophisticated, and you get queries that you just wouldn’t get in the United States. For example, we get queries like, “Hi Baidu, how are you? I ate noodles at a corner store last week and they were delicious. Do you think they’re on sale this weekend?” That’s the query.
  
  WSJ: You can process that?
  
  Ng: If they speak clearly, we can do the transcription fairly well and then I think we make a good attempt at answering. Honestly, the funniest ones are schoolchildren asking questions like: “Two trains leave at 5 o’ clock, one from …” That one we’ve made a smaller investment in, dealing with the children’s homework. In China, a lot of users’ first computational device is their smartphone, they’ve never owned a laptop, never owned a PC. It’s a challenge and an opportunity.
  
  WSJ: You have the Baidu Eye, a head-mounted device similar to Google Glass. How is that project going?
  
  Ng: Baidu Eye is not a product, it’s a research exploration. It might be more likely that we’ll find one or two verticals where it adds a lot of value and we’d recommend you wear Baidu Eye when you engage in certain activities, such as shopping or visiting museums. Building something that works for everything 24/7 – that is challenging.
  
  WSJ: What about the self-driving car project? We know Baidu has partnered with BMW BMW.XE +0.90% on that.
  
  Ng: That’s another research exploration. Building self-driving cars is really hard. I think making it achieve high levels of safety is challenging. It’s a relatively early project. Building something that is safe enough to drive hundreds of thousands of miles, including roads that you haven’t seen before, roads that you don’t have a map of, roads where someone might have started to do construction just 10 minutes ago, that is hard.
  
  WSJ: How does working at Baidu compare to your experience at Google?
  
  Ng: Google is a great company, I don’t want to compare against Google specifically but I can speak about Baidu. Baidu is an incredibly nimble company. Stuff just moves, decisions get made incredibly quickly. There’s a willingness to try things out to see if they work. I think that’s why Baidu, as far as I can tell, has shipped more deep-learning products than any other company, including things at the heart of our business model. Our advertising today is powered by deep learning.
  
  WSJ: Who’s at the forefront of deep learning?
  
  Ng: There are a lot of deep-learning startups. Unfortunately, deep learning is so hot today that there are startups that call themselves deep learning using a somewhat generous interpretation. It’s creating tons of value for users and for companies, but there’s also a lot of hype. We tend to say deep learning is loosely a simulation of the brain. That sound bite is so easy for all of us to use that it sometimes causes people to over-extrapolate to what deep learning is. The reality is it’s really very different than the brain. We barely (even) know what the human brain does.
  
  WSJ: For all of Baidu’s achievements, it still has to operate within China’s constraints. How do you see your work and whether its potential might be limited?
  
  Ng: Obviously, before I joined Baidu this was something I thought about carefully. I think that today, Baidu has done more than any other organization to open the information horizon of the Chinese people. When Baidu operates in China, we obey Chinese law. When we operate in Brazil, which we also do, we obey Brazil’s law. When we operate in the U.S. and have an office here, we obey U.S. law. When a user searches on Baidu, it’s clear that they would like to see a full set of results. I’m comfortable with what Baidu is doing today and I’m excited to continue to improve service to users in China and worldwide.
  
  - Rolfe Winkler contributed to this article.
  

 

[TaiShang]

Baidu sets up deep learning lab
Xinhua, March 3, 2017

Baidu sets up deep learning lab [Baidu]

China's Internet giant Baidu inaugurated a national deep learning technology lab in Beijing on Thursday to boost research in the field.

The national engineering lab of deep learning technology and application will be led by Baidu and built in cooperation with partner researchers and institutes, said Baidu in a press release.

The goal of the lab is to integrate resources from government, industry leaders and academic communities to improve the country's overall research capability and applications in the area of deep learning and artificial intelligence, it said.

The lab will strengthen innovation capability to take the lead in the development of the sector, said Sun Wei, an official with the National Development and Reform Commission.

The lab will focus research on areas such as visual perception, speech recognition, biometrics recognition and human-machine interaction.