This repository provides a multi-mode and multi-speaker expressive speech synthesis framework, including multi-attentive Tacotron, DurIAN, Non-attentive Tacotron.

The framework also includes various deep learning architectures such as Global Style Token (GST), Variational Autoencoder (VAE), and Gaussian Mixture Variational Autoencoder (GMVAE), and X-vectors for building prosody encoder.

• Only provides kernel model files, not including data prepared scripts, training scripts and synthesis scripts
• You can reference ExpressiveTacotron for more training scripts

• Global Style Token (GST)
• Variational Autoencoder (VAE)
• Gaussian Mixture VAE (GMVAE)
• X-vectors

• Tacotron2
• ForwardAttention
• DurIAN
• Non-attentive Tacotron
• GMMv2 Attention
• Dynamic Convolution Attention (Todo)

• Non-attentive Tacotron: duration stacked convolution layers are concatenated with encoder outputs

This implementation uses code from the following repos: NVIDIA, ESPNet, ERISHA, ForwardAttention