Skip to content

Latest commit

 

History

History
72 lines (58 loc) · 2.5 KB

README.md

File metadata and controls

72 lines (58 loc) · 2.5 KB
Raw

Tacotron 2

A PyTorch implementation of Tacotron2, described in Natural TTS Synthesis By Conditioning Wavenet On Mel Spectrogram Predictions, an end-to-end text-to-speech(TTS) neural network architecture, which directly converts character text sequence to speech.

Install

  • Python3.6+ (Recommend Anaconda)
  • PyTorch 0.4.1+
  • pip install -r requirements.txt
  • If you want to run egs/ljspeech/run.sh, download LJ Speech Dataset for free.

Usage

Quick start

$ cd egs/ljspeech
# Modify wav_dir to your LJ Speech dir
$ bash run.sh

That's all.

You can change parameter by $ bash run.sh --parameter_name parameter_value, egs, $ bash run.sh --stage 2. See parameter name in egs/ljspeech/run.sh before . utils/parse_options.sh.

Workflow

Workflow of egs/ljspeech/run.sh:

  • Stage 1: Training
  • Stage 2: Synthesising

More detail

egs/ljspeech/run.sh provide example usage.

# Set PATH and PYTHONPATH
$ cd egs/ljspeech/; . ./path.sh
# Train:
$ train.py -h
# Synthesis audio:
$ synthesis.py -h

How to visualize loss?

If you want to visualize your loss, you can use visdom to do that:

  1. Open a new terminal in your remote server (recommend tmux) and run $ visdom
  2. Open a new terminal and run $ bash run.sh --visdom 1 --visdom_id "<any-string>" or $ train.py ... --visdom 1 --vidsdom_id "<any-string>"
  3. Open your browser and type <your-remote-server-ip>:8097, egs, 127.0.0.1:8097
  4. In visdom website, chose <any-string> in Environment to see your loss loss

How to resume training?

$ bash run.sh --continue_from <model-path>

How to use multi-GPU?

Use comma separated gpu-id sequence, such as:

$ bash run.sh --id "0,1"

How to solve out of memory?

  • When happened in training, try to reduce batch_size or use more GPU. $ bash run.sh --batch_size <lower-value> or $ bash run.sh --id "0,1".

Reference and Resource

NOTE

This is a work in progress and any contribution is welcome (dev branch is main development branch).

I implement feature prediction network + Griffin-Lim to synthesis speech now.

Attention and synthesised audio on 37k iterations: attn spec