Introduction

This repository allows us to build a website, which converts the "text" data into the "speech" data using two deeplearning models: WaveGlow and Tacotron2. Since the computation requires GPU to get a result in a reasonable amount of time, my demo was created using RTX TITAN.

Demo

Introduction

website_intro.mp4

Sample process

It takes a little bit to process the text data, so please be patient.

website_process.mp4

Website

Applications

There are some applications used to create this website as shown below. Thanks to docker (docker-compose), I was able to create three servers (containers) easily, being categorized as a web server, a python server, and a database.

Overview

The whole process of text2speech is illustrated below. As for #2 and #3, if the input text is already processed before, it will just output the audio from the dataset. Of course, if there is no expected speech in the dataset, it processes the text to obtain the speech using the python server.

How to use

Initialization

After cloning this repository, please navigate to the main folder and run the code below with bash.

bash -c "$(curl -fsSL https://raw.githubusercontent.com/shinshoji01/text2speech-website/main/bin/install.sh)"

Build a website

1. docker-compose up -d --build
2. go to http://localhost:8085/index.php.