Lower Sorbian voice data for MaryTTS

Lower Sorbian speech database for TTS voices in MaryTTS, recorded in 2022 at the Sorbian Institute.

Format

The audio data is provided in a single FLAC file, recorded at 44.1 kHz sampling frequency with 16 bit per sample.

The textual data is provided in a single YAML file. This file is a list of utterances, each of which contains

• a prompt code (file basename),
• the utterance text,
• utterance start and end times (in seconds) in the FLAC file,
• optionally, the phonetic segments, each of which has
  • a label (based on SAMPA, _ denotes silence), and
  • its duration (in seconds)

The phonetic segmentation has been prepared using the Kaldi-based Montreal Forced Aligner.

Downloading the data

Use the links on the releases page, or run the unpackData Gradle task (see below).

Converting the data

For convenience, the utterances for each subset can be be extracted from the YAML and FLAC files using simple commands to run Gradle tasks. After cloning or downloading and unpacking this repository, run ./gradlew tasks (or gradlew tasks on Windows) for details.

Prerequisites

You will need Java to run the tasks. Extracting the utterances to WAV files also requires SoX to be installed.

Assembling the data

The data processing is delegated to Gradle and the Kaldi MFA and FLAML plugins.

Run ./gradlew unpackData to only download and extract all data.

Run ./gradlew assemble to download, extract, and force-align all data, and prepare it for distribution.

Copyright and license

Copyright 2022 Sorbian Institute.

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.