LIBRITTS SYLLABLES CORPUS FOR PALS0039

The following directories contain (maximum-onset) syllables extracted from the LibriTTS corpus using phonetic forced-alignments from the LibriTTSLabel resource.

The audio files are extracted from the clean subsets of the corpus and the train files are from the clean-100 subset.

Description of files

• The audio files are organised into three sub-directories: dev, test, and train.
• Each audio file contains a single syllable extracted from the continuous speech.
• The file name describes where each syllable is found (speaker, utterance, and word origins) and the phonetic description in stressed Arpabet format.

An example of the information contained in each filename:

1272_128104_000001_000000_0002_jolly_JH-AA1.wav

This contains the following information separated by underscores:

• 1272_128104_000001_000000 --> This is the utterance id in the LibriTTS corpus, of which 1272 is the speaker id.
• 0002 --> This is the syllable index (i.e. the third syllable extracted from this utterance).
• jolly --> The surface form of the word in the utterance.
• JH-AA1 --> The phonetic transcription for this syllable (i.e. it is the first syllable in "jolly")

For convenience in downloading, organising, selecting and partitioning the data, the following files contain lists of the filenames in each of the sub-directories dev.txt, test.txt, and train.txt.

You are not required to use the complete dataset, however, all selections should be described and motivated appropriately.

If you prefer to download the entire corpus to local storage you may use the gzipped tar files: dev.tar.gz, test.tar.gz, and train.tar.gz.