Natural TTS Synthesis by Conditioning WaveNet on Mel Spectrogram Predictions

Description

An implementation of the Tacotron 2 model from the Natural TTS Synthesis by Conditioning WaveNet on Mel Spectrogram Predictions paper. It contains some modifications to the original paper such as:

• Removed stop token prediction
• Replaced the WaveNet vocoder with the Griffin Lim algorithm used in the original Tacotron model

Other

The repo contains a PDF written in Romanian. This PDF is the description of the project since this was used in a class homework. Also, because I lacked the necessary resources to train the model, the demo uses the model provided by NVIDIA and my own implementation of the Griffin-Lim algorithm.

Requierments

• PyTorch
• Librosa
• Pandas

Credits

• https://github.com/A-Jacobson/tacotron2
• https://github.com/NVIDIA/tacotron2