TTS performance in relation to other solutions.

[erogol]

Above you can see the figure comparing TTS with the other commercial and non-commercial solutions. Comparison has been made by asking ~500 human participants after they listen to the same article from all these solutions.

TLDR; It is not the best (yet) but good to see an open-source project is comparable with big commercial solutions!

Models are defined as follows:
Mozilla TTS Nancy2: Tacotron2 + WaveRNN
Mozilla TTS Nancy: Tacotron
Mozilla TTS LJSpeech: Tacotron

Note that .<name> are real people reading the same article.

Here is the article https://soundcloud.com/user-565970875/commonvoice-1

[OswaldoBornemann]

Glad to see that.! Could we train TTS (tacotron2 version) using dev-tacotron2 branch?

[PetrochukM]

What article did you ask the 500 participants to read?

[erogol]

Added