jackz314 / speech-synthesizer

File to speech

Speech Synthesizer

A GUI application that does end-to-end human-realistic text to speech.

Overview

The app uses espnet for most of the heavy lifting (text -> spectrogram generation -> wave generation), I added some text preprocessing steps that improves support for Chinese, as well as automatic language identification (with langdetect) for ease of use.

In addition, the app supports reading from files, including a wide range of formats thanks to the Calibre integration. It also outputs a subtitle and lyrics file synced for each sentence.

The GUI is implemented with PySide2 (Qt for Python), and the backend-frontend communication is implemented with pyzmq.

Installation

To run locally, clone the repository and install the requirements:

pip install -r requirements.txt

Then run ui.py to start the app (python ./ui.py).

Alternatively, download zip from releases, extract it and run ui.exe, currently only Windows has pre-built releases.

OS Support

Everything I used is cross-platform, this has been roughly tested on both Windows and Ubuntu 20.10, but it should also work on other platforms.

Deployment

Deployment of this app is done with PyInstaller, it freezes the app and its dependencies into one huge (due to dependency on large libraries) directory.

To deploy, install pyinstaller (pip install pyinstaller) and run:

pyinstaller ui.spec