F0-DCTTS (F0 Deep Convolutional TTS)

Description

• DCTTS with F0

Prerequisite

• python 3.7
• pytorch 1.3
• pysptk
• librosa, scipy, tqdm, tensorboardX

Dataset

• LJ Speech 1.1
• KSS, Korean female single speaker speech dataset.

Samples

• samples

Usage

1. Download the above dataset and modify the path in config.py. And then run the below command.

   python prepro.py
   

2. The baseline DCTTS needs to train 100k+ steps

   python train.py <gpu_id>
   

3. After training the baseline, you can train F0-DCTTS. Change "f0_mode=True" and "pretrained_path=..." in config.py. And then run the below command one more.

   python train.py <gpu_id>
   

4. You can synthesize some speech with f0. You can control using "f0_factor=..." in config.py.

   python synthesize.py <gpu_id>
   

Notes

• This method is easy and simple, but verrrrrrrrrry naive approach.
• In this code, I removed SSRN. Thus, you need another mel2wav vocoder. I recommend WaveGlow or Parallel WaveGAN.