EdenMelaku/Amharic-TTS-with-Tacotron2

# Amharic Text to speech with Tacotron2 This repo is text to speech implemented on **Tacotron2** paper Natural TTS Synthesis By Conditioning Wavenet On Mel Spectrogram Predictions Basic Model implementation is adobted from NVIDIA'S tacotron2 implemetation and [NVIDIA'S waveglow implementation] (github.com). it is implemented by pytorch and it is only supported with GPU. I am inspired by Tacotron implementation by Keith Ito. ## audio samples generated by the model (Tacotron2 + waveglow) We have trained 2 different samples both found in this repo. The difference is how big of training data we have used the initial one was 11 hours and another 27 hours, We have also learned alot from the 11 hours data and made some modifications on how we trained the model. * 11Hrs - Generated Audio Samples/First model * 27Hrs - Generated Audio Samples/Latest model Learn more about the experiments here Presentation slide : docs.google.com