Baidu Inc (NASDAQ:BIDU) Announced That Its Deep Speech Can Recognize Voice Even In Noisy Environments

Andrew Ng joined Baidu Inc (NASDAQ:BIDU), a Chinese internet company, back in May 2014 as chief scientist. He was a little worried about the projects that he and his team would be working on in the newly built labs of Sunnyvale, California. To the surprise of everyone, the company designed a speech recognition that is much better than any other such program available in the market.

Baidu Inc (NASDAQ:BIDU) is also known as the Google of China. The company posted those results which the professors of Stanford, cofounder of Coursera and researchers of Google had in mind. Andrew Ng, together with his research team members, published a paper on December 18, 2014, on the library Website of Cornell University. Andrew’s team, which was being led by Awni Hannun, a research scientist, posted that they have designed a new speech recognition method. It is important to mention here that speech recognition is being used in the voice search of Google and Siri program of Apple.

The deep speech method of recognition offered by Baidu is much better than that of Apple’s and Google’s, according to the claims of Andrew Ng. The method designed by Andrew will be able to measure and filter out the errors in speech recognition; meaning that the program will be able to work well in the noisy environments, for instance, in parks, crowds or cars. If that is indeed true, this method is of much importance in the real world use.

According to the report of Andrew Ng, Baidu Inc’s (NASDAQ:BIDU) deep speech, when tested in the noisy environment, performed much better than the other similar systems, which included Bing Speech of Microsoft, Speech API of Google and Dictation of Apple. In terms of error rates in words, Deep Speech was 10 percent better on average than all the other systems.

Carnegie Mellon University’s assistant research professor, Ian Lane, who teaches engineering students, remarked that this breakthrough can completely change the future of speech recognition. Apple, Google and Microsoft have avoided commenting on this news.

Baidu’s Deep Speech, like all other recognition programs, is based on Artificial intelligence’s deep learning branch. This software imitates the neuron activities that happen in the brain when a person is thinking or learning something. The activities are then translated in the digital pattern form.

The research team of Baidu Inc (NASDAQ:BIDU) contacted 9,600 people and gathered around 7,000 hours of talking and speech in the quiet backgrounds. Sometimes, the speakers were requested to put on hands free while loud noise was being played in the background so as to change the inflections and pitch. The team then used what is called the principle of superposition in physics to add around 15 different types of noise, which included noise in the car, subway, restaurants and parks. This technique amplified the data to around 100,000 hours. The system had been then run in order for the programs to recognize speech amidst all this noise.

The stocks of Baidu Inc (NASDAQ:BIDU) saw an increase of 3.59 percent as soon as the news got out.