19 Jan
Google Shows Off A.I.-Powered Voice Capabilities Google Shows Off A.I.-Powered Voice Capabilities

Google Shows Off A.I.-Powered Voice Capabilities

Written by 

If you’ve spent much time interacting with voice assistants like Siri and Alexa, you know that text-to-speech isn’t an exact science yet. Digital diction tends to sound more robotic than human voices, although technology has improved things in recent years.

Thanks to Google, things will only continue to get better. The tech giant recently published a paper that described its upcoming Tacotron 2, an advanced way to electronically deliver speech that achieves a mean opinion score (MOS) of 4.53 compared to the 4.58 score for recorded human speech. The MOS score measures how realistic something sounds.

The technology represents the sound through a spectrogram, which is put through Google’s WaveNet algorithm to translate it into eerily realistic speech. As Google demonstrates, Tacotron 2 has the ability to make subtle differentiations in English words, such as realizing the varying uses of the nouns “desert” and “present” and the verbs that are spelled exactly the same but pronounced differently.

The Google team was vague about how far along they are in the development of this new system, but they did leave clues for developers to decipher. One notable observation from the paper was that the sound samples provided had a filename that included either “gen” or “gt.”

Rate this item
(0 votes)
Read 3188 times Last modified on Wednesday, 03 January 2018 03:51
Stephanie Faris

Stephanie is a freelance writer and novelist whose work has appeared on,, the Intuit Small Business Blog, and many others. She is the Simon & Schuster author of 8 children's books, including the Piper Morgan chapter book series, 25 Roses, and 30 Days of No Gossip.

Login to post comments
Home Our Blog Google Shows Off A.I.-Powered Voice Capabilities

Newsletter Signup

Live support

Available Monday - Friday, 9 AM - 5 PM EST

Connect with us

Netributor Main Offices