Project AI❤dol Public English Dataset

Due to a severe lack of publicly available English SVS data sung by a female singer, I have decided to publish my own corpus. This is a selection of the full dataset; only English songs have been included, and only the very best singing data has been selected. The voice provider is not a native English speaker, however the pronunciation should be good enough. She is also not a professional singer, however the singing should be of passable quality nonetheless.

This dataset uses ARPABET phonemes, including the extended phonemes ax (schwa), dx (tap) and en (syllabic n). It also features the extra phonemes EP (exhale), GS (glottal stop) and vf (vocal fry/creaky voice). (The ARPABET symbol q was not used for glottal stops, due to a conflict with Chinese phoneme systems that are based on pinyin.)

Total duration of data: 1 hour and 45 minutes.

General Terms of Use

• Please refer to the LICENSE for more information before using this dataset.
• This dataset is primarily intended for parallel training. Please do not publicly release any model featuring this voice without prior permission.
• Please give proper attribution to Lotte V (or @lottev1991) as the creator of this dataset (see LICENSE).
• You are allowed to use this dataset for research purposes.
• You are allowed to release customized labels for this dataset without prior permission, as long as proper attribution is given.
• You are allowed to use the labels for your own singing data; please keep in mind that you might need to manually adjust them. Please make sure that proper attribution is given.
• You are allowed to use this dataset for pitch training for your own SVS model, as long as proper attribution is given.
• Please do not use this dataset with voice changers, such as RVC etc.
• Please do not use this dataset to train any commercial models (see LICENSE).
  • However, any models that are trained with this dataset are allowed to be used in commercial creative projects.
    • You may have to ask permission to the model trainer to use their specific model in commercial projects.
• Please do not use this dataset for any illegal purposes.
• Please do not use this dataset for training unauthorized vocals (such as celebrity voices, etc. (unless they have explicitly and officially authorized their voice to be used for AI model training)).
• This dataset should never be used in association with any religious organizations.
  • However, any model resulting from this dataset is allowed to be used in music with religious themes, as long as the above is kept in mind.