diSpeech, a corpus of synthesized speech for disentanglement purposes

diSpeech is a corpus of speech synthesized with the Klatt synthesizer to generate datasets for speech disentanglement purposes. The purpose of disentanglement is to automatically extract and separate the attributes constituting the speech signal.
The Klatt synthesizer is used to generate phonemes.
The first version, constrained to vowels synthesized with 5 generative factors relying on pitch and formants, can be used in experiments that address fundamental but still misunderstood aspects of speech disentanglement.

Available sur github.com/Orange-OpenSource/diSpeech.


Published on