voice-anna-librivox

A female Dutch HMM voice for MaryTTS built from LibriVox data: Een Nagelaten Bekentenis and Majoor Frans

Sentence-split text files and original silence-segmented wav files were semi-automatically produced and can be found here

A subset was selected based on average loudness, with sound files normalized to -1 db using:

for f in wav-orig/* ; do sox --norm=-1 $f wav/`basename $f` ; done

Prerequisites

On Debian-based Linux (including Ubuntu), do

$ sudo apt-get install default-jdk sox speech-tools

For the kaldi-based forced aligner, install the following: (see (https://github.com/marytts/gradle-marytts-kaldi-mfa-plugin))

$ sudo apt-get install libatlas3-base

accordingly.

Initializing the build

Do

$ ./gradlew legacyInit

Building the voice

To assemble, test, and package the voice for another MaryTTS installation, do

$ ./gradlew build

The packaged files will be placed under build/distributions. Copy the zip file and xml descriptor into your MaryTTS installation's download directory, then run the marytts-component-installer to install the voice.

Running the voice

To build the voice and run it in an ad-hoc MaryTTS server, do

$ ./gradlew run

Then, go to http://localhost:59125.