Registration was successful!
Please follow the link from the email sent to

Google Breaks Ground With Best Artificial Intelligence Speech Generator Yet

© AP Photo / Virginia MayoFILE - In this March 23, 2010, file photo, the Google logo is seen at the Google headquarters in Brussels.
FILE - In this March 23, 2010, file photo, the Google logo is seen at the Google headquarters in Brussels. - Sputnik International
A UK-based Google office is claiming that its newly developed speech generating system performs with 50 percent more proficiency than existing technologies.

The DeepMind unit, which Google bought in 2014 for roughly $533 million, develops supercomputers and artificial intelligence (AI). One of the AI programs DeepMind has developed is Wavenet, designed to mimic human speech. 

Pokemon Go - Sputnik International
Hello Monsters, Goodbye Privacy! Google Maps Timeline and Pokemon Go Pally Up

In blind tests human listeners indicated that Wavenet was the most natural sounding text-to-speech (TTS) program, after hearing samples from different programs in English and Mandarin Chinese. TTS programs continue to struggle to sound like natural speech, and Wavenet, while increasingly similar, does not yet sound just like actual human speech.

Wavenet simulates certain brain functions by using what in artificial intelligence is called a "neural network." DeepMind has reported that Wavenet is not currently available to purchase as the processing power required to sample an audio signal some 16,000 times per second remains impractical at the consumer level. 

Once the signal is sampled, it must predict what each soundwave should look like based on prior samples. In a blog post, DeepMind researchers admit that the undertaking "is a clearly challenging task." 

This combination of file photos shows Democratic presidential candidate Hillary Clinton(R) speaking at New York University in New York on July 24, 2015 and US Republican presidential candidate Donald Trump exiting the New York Supreme Court after morning jury duty on August 17, 2015 in New York - Sputnik International
Can Google Tip the Scales on the US Presidential Election Without Us Knowing?

As people increasingly use speech to interface with technology, from cars to mobile phones, other tech companies are taking note of DeepMind’s advancements. Apple, Microsoft, Google-parent Alphabet, and Amazon already communicate using speech-based digital personal assistants.

At a London developer conference last week, Google Play’s international director, Mike Bennett, revealed that speech commands make up 20 percent of Google searches, as opposed to text searches.

Google acknowledged that it has used DeepMind to drop the power demands of its data centers by 40 percent, but has not said much about how the program has impacted the company commercially, if at all. The company has also stated that the AI has made "substantial improvements to a set of services from YouTube and Google Play to Google’s advertising products."

Before developing Wavenet, Google also made strides in AI with the AlphaGo program, which was able to beat Lee Sedol, the world’s best player of strategy game Go. 

To participate in the discussion
log in or register
Заголовок открываемого материала