Скачать или смотреть Sanskrit TTS using Tacotron2, WaveGlow and Transfer Learning

We have designed a Sanskrit TTS with very little speech corpus, on a deep learning framework, using transfer learning. We request you kindly listen to the samples of the synthesized speech and evaluate it. The Sanskrit speech data was downloaded and prepared in the required format by Ms. Gangotri and Ms. Shridevi, BE interns in our lab. The training and tuning of Tacotron2, with all the other refinements needed to get the TTS going, was done by Mr. Ankur Debnath, a bright and committed M Tech student, as one of the parts of his final project. We intend to expand the work and come out with a very high-quality TTS for Sanskrit, the oldest (next only to Tamil) language. This is our work presented at INDICON 2020.
For more details, read our paper:
https://www.researchgate.net/publicat...
You can find the code and the data at https://github.com/AnkurDebnath35/Low...
If you use our data or code, please cite our paper as follows:
D Ankur, N Gangotri, SP Shridevi, AG Ramakrishnan, “Low-Resource End-to-end Sanskrit TTS using Tacotron2, WaveGlow and Transfer Learning,” Proc. IEEE INDICON 2020.