Implementation of TTS with combination of Tacotron2 and HiFi-GAN for Mandarin TTS.
In order to inference, we need to download pre-trained tacotraon2 model for mandarin, and place in the root path. Then, we can run
infer_tacotron2_hifigan.py to get TTS result. We can alter the input text by editting variablle
text in the
infer_tacotron2_hifigan.py. Then the result will be saved in the root path named as
The pre-trained model of HiFi-GAN has been placed in the
LJ_FT_T2_V3, which is trained by LJSppech and fine-tuned with Tacotron2. You can find more pre-trained model from original HiFi-GAN repo with different size and parameters. If you want to try different models or train your own model, please do remember to alter variables in
infer_tacotron2_hifigan.py to change the path of HiFi-GAN model.