-
Notifications
You must be signed in to change notification settings - Fork 4.5k
MOS Results
Eren Gölge edited this page Mar 5, 2021
·
2 revisions
Based on our latest MOS study, 🐸 TTS is able to achieve on par performance with any other solution. However, it is also shown that the quality of the dataset is an important part of 🐸 TTS. Therefore, it is worth to find a good resource or even record your own dataset, if you like to peak the 🐸 TTS performance.
- Judy Wave1: Tacotron + WaveRNN
- Judy Wave2: Tacotron2 + WaveRNN
- Judy GL1 : Tacotron + Griffin Lim
- Judy GL2 : Tacotron2 + Griffin Lim
- TTS Nancy: Tacotron + Griffin Lim
- TTS Nancy2: Tacotron2 + WaveRNN
- TTS LJSpeech: Tacotron + GriffinLim
(Judy is an 25 hours long internal dataset)
(.Jofish .Abe and .Janice are real human voices)