Can you tell the difference between a real human voice and Google's new AI voice?

Can you tell the difference between a real human voice and Google's new AI voice?

Google has developed a new AI-based text-to-speech system - the Tacotron 2 - that sounds indistinguishable from the voice of a real human, at least that is what Google claims.

To be perfectly exact, the Tacotron 2 system received a mean opinion score (MOS) of 4.53 comparable to a MOS of 4.58 for a professionally recorded speech.

The new system does not sound robotic or digitized in any easily noticeable way, and it can even tell the correct pronunciation of words depending on the semantics. It can also deal with some slight typing errors and can do things like tongue-twisters.


We tried it and we could not tell the difference between a human voice and this new system. So let's play a fun game... below you will find four samples recorded twice. One of the recordings is the Tacotron 2 AI voice and the second one is the professional human narrator. Can you tell which one is which? (We have the answers at the end of this article.)

1. Which one is real voice?


A:
 

B:


2. Which one is real voice?


C:


D:


3. Which one is real voice?


E:


F:


4. Which one is real voice?


G:


H:


Could you tell the difference?

*We were able to hear the difference after listening to the samples a few times, but not initially.

Here are the correct answers for which one is which.

Answers:

A - Human voice
B - Tacotron 2 AI voice

C - Tacotron 2 AI voice
D - Human voice

E - Tacotron 2 AI voice
F - Human voice

G - Human voice
H - Tacotron 2 AI voice

FEATURED VIDEO

10 Comments

1. Soundjudgment

Posts: 370; Member since: Oct 10, 2016

Get it some real challenges: "How Now Brown Cow." "Unique New York, Unique New York, Unique New York." "su·per·ca·li·fra·gil·is·tic·ex·pi·a·li​·do·cious"

2. Phonehex

Posts: 700; Member since: Feb 16, 2016

Pretty close !

3. peace247 unregistered

Amazing

4. Behyjoon

Posts: 39; Member since: Oct 19, 2016

That was close but i could figure out all of them still needs work

5. cmdacos

Posts: 3661; Member since: Nov 01, 2016

I got a couple wrong but largely because the real person doesn't have a natural speech pattern either.

6. Whitedot

Posts: 632; Member since: Sep 26, 2017

I found both on test 1 pretty human-sound like while Ai is more obvious on remaining tests.

7. warrenellis93

Posts: 523; Member since: Jul 21, 2011

They both sound like robots

8. Leo_MC

Posts: 5915; Member since: Dec 02, 2011

It's easy to spot the AI when you listen to the rhythm of the phrase.

9. drifter77

Posts: 397; Member since: Jun 12, 2015

#2 was a little tricky.

* Some comments have been hidden, because they don't meet the discussions rules.

Latest Stories