TTS performance in relation to other solutions.
erogol opened this issue · 3 comments
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
Glad to see that.! Could we train TTS (tacotron2 version) using dev-tacotron2 branch?
What article did you ask the 500 participants to read?
Added