# WaveRNNTTS

**Repository Path**: yanlang0123/WaveRNNTTS

## Basic Information

- **Project Name**: WaveRNNTTS
- **Description**: WaveRNN Vocoder + TTS
- **Primary Language**: Unknown
- **License**: MIT
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 1
- **Created**: 2021-12-03
- **Last Updated**: 2022-05-06

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# WaveRNN

##### (Update: Vanilla Tacotron One TTS system just implemented - more coming soon!)

![Tacotron with WaveRNN diagrams](assets/tacotron_wavernn.png)

Pytorch implementation of Deepmind's WaveRNN model from [Efficient Neural Audio Synthesis](https://arxiv.org/abs/1802.08435v1)

# Installation

Ensure you have:

* Python >= 3.6
* [Pytorch 1 with CUDA](https://pytorch.org/)

Then install the rest with pip:

> pip install -r requirements.txt

# How to Use

### Quick Start

If you want to use TTS functionality immediately you can simply use:

> python quick_start.py

This will generate everything in the default sentences.txt file and output to a new 'quick_start' folder where you can playback the wav files and take a look at the attention plots

You can also use that script to generate custom tts sentences and/or use '-u' to generate unbatched (better audio quality):

> python quick_start.py -u --input_text "What will happen if I run this command?"


### Training your own Models
![Attenion and Mel Training GIF](assets/training_viz.gif)

Download the [LJSpeech](https://keithito.com/LJ-Speech-Dataset/) Dataset.

Edit **hparams.py**, point **wav_path** to your dataset and run:

> python preprocess.py

or use preprocess.py --path to point directly to the dataset
___

Here's my recommendation on what order to run things:

1 - Train Tacotron with:

> python train_tacotron.py

2 - You can leave that finish training or at any point you can use:

> python train_tacotron.py --force_gta

this will force tactron to create a GTA dataset even if it hasn't finish training.

3 - Train WaveRNN with:

> python train_wavernn.py --gta

NB: You can always just run train_wavernn.py without --gta if you're not interested in TTS.

4 - Generate Sentences with both models using:

> python gen_tacotron.py wavernn

this will generate default sentences. If you want generate custom sentences you can use

> python gen_tacotron.py --input_text "this is whatever you want it to be" wavernn

And finally, you can always use --help on any of those scripts to see what options are available :)



# Samples

[Can be found here.](https://fatchord.github.io/model_outputs/)

# Pretrained Models

Currently there are two pretrained models available in the /pretrained/ folder':

Both are trained on LJSpeech

* WaveRNN (Mixture of Logistics output) trained to 800k steps
* Tacotron trained to 180k steps

____

### References

* [Efficient Neural Audio Synthesis](https://arxiv.org/abs/1802.08435v1)
* [Tacotron: Towards End-to-End Speech Synthesis](https://arxiv.org/abs/1703.10135)
* [Natural TTS Synthesis by Conditioning WaveNet on Mel Spectrogram Predictions](https://arxiv.org/abs/1712.05884)

### Acknowlegements

* [https://github.com/keithito/tacotron](https://github.com/keithito/tacotron)
* [https://github.com/r9y9/wavenet_vocoder](https://github.com/r9y9/wavenet_vocoder)
* Special thanks to github users [G-Wang](https://github.com/G-Wang), [geneing](https://github.com/geneing) & [erogol](https://github.com/erogol)
如何使用
快速开始
如果您想立即使用 TTS 功能，您可以简单地使用：

python quick_start.py

这将在默认的 sentence.txt 文件中生成所有内容并输出到一个新的“quick_start”文件夹，您可以在其中播放 wav 文件并查看注意力图

您还可以使用该脚本生成自定义 tts 句子和/或使用“-u”生成未批处理（更好的音频质量）：

python quick_start.py -u --input_text "如果我运行这个命令会发生什么？"

训练你自己的模型
注意和梅尔训练 GIF

下载LJSpeech数据集。

编辑hparams.py，将wav_path指向您的数据集并运行：

蟒蛇预处理.py

或使用 preprocess.py --path 直接指向数据集

这是我对运行顺序的建议：

1 - 训练 Tacotron：

python train_tacotron.py

2 - 您可以离开完成培训或在任何时候使用：

python train_tacotron.py --force_gta

这将迫使 tactron 创建一个 GTA 数据集，即使它还没有完成训练。

3 - 训练 WaveRNN：

python train_wavernn.py --gta

注意：如果你对 TTS 不感兴趣，你可以在没有 --gta 的情况下运行 train_wavernn.py。

4 - 使用两种模型生成句子：

python gen_tacotron.py wavernn

这将生成默认句子。如果你想生成自定义句子，你可以使用

python gen_tacotron.py --input_text "这就是你想要的" wavernn

最后，您始终可以在任何这些脚本上使用 --help 来查看可用的选项:)