Gradle MaryTTS voicebuilding plugin

Use this plugin to build new unitselection voices for MaryTTS. It's a replacement for the legacy VoiceImportTools.

Applying the plugin

Add this at the top of your build.gradle file:

plugins {
    id 'de.dfki.mary.voicebuilding-legacy' version '5.4'
}

For details, see https://plugins.gradle.org/plugin/de.dfki.mary.voicebuilding-legacy.

Note that Gradle v6.2 or higher is required.

Prerequisites

Required third-party software

Java 8 or higher is required.

SoX and Edinburgh Speech Tools must be installed and on the PATH.

Mac OSX

With Homebrew, just run

brew install sox speech-tools

Linux

On Debian-based systems, just run

sudo apt install sox speech-tools

Project layout

In your project directory, place the source audio, text, and label files under your build directory like this:

build
├── lab
│   ├── utt0001.lab
│   ├── utt0002.lab
│   ├── utt0003.lab
│   ├── utt0004.lab
│   └── utt0005.lab
├── text
│   ├── utt0001.txt
│   ├── utt0002.txt
│   ├── utt0003.txt
│   ├── utt0004.txt
│   └── utt0005.txt
└── resources
    └── data
        ├── utt0001.wav
        ├── utt0002.wav
        ├── utt0003.wav
        ├── utt0004.wav
        └── utt0005.wav

The audio location can be easily customized, by configuring the layout in the build.gradle like this:

// wav files under "$projectDir/wav"
wav.srcDir = file('wav')

Configuring the voice

Crucial details of the voice to be built are configured in the marytts.voice extension in the build.gradle file. Without any customization, the defaults are:

marytts {
    voice {
        name = 'my_voice'
        gender = 'female'
        language = 'en'
        region = 'US'
        domain = 'general'
        type = 'unit selection'
        description = 'A female English unit selection voice'
        samplingRate = 16000
        license {
            name = 'Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International'
            shortName = 'CC BY-NC-SA 4.0'
            url = 'http://creativecommons.org/licenses/by-nc-sa/4.0/'
        }
    }
}

Any of these can (and should) be changed as needed, according to the dataset used to build the voice. The default values do not need to be specified, and the description is generated using the values of gender, language, region, and type.

Building the voice

To assemble and test the voice, run

./gradlew build

Afterwards, the packaged voice component (and its XML descriptor) will be found under build/distributions. These files can be installed in a MaryTTS v5.2 instance, by placing them in the download directory and running the MaryTTS Component Installer GUI.