Implement the Centaur model

NVIDIA / OpenSeq2Seq

Toolkit for efficient experimentation with Speech Recognition, Text2Speech and NLP

https://nvidia.github.io/OpenSeq2Seq

Apache License 2.0

1.54k stars 369 forks source link

Closed rebryk closed 5 years ago

rebryk commented 5 years ago

Centaur is TTS model:

The encoder consists of convolutional layers.
The decoder consists of pre-net and modified transformer blocks followed by convolutional layers.