Silasimo/Project-AIdol-Segmented

This dataset contains segmented audio and customized annotations derived from the Project AI❤dol Public English Dataset by Lotte V (@lottev1991). The original dataset is licensed under cc-by-sa-4.0, and thus we do the same with this segmented version. To learn more about the terms of use, please refer to the original dataset license. This dataset has been constructed as a benchmark for singing-oriented forced alignment in our SynthGT project. We manually decide on segmentation points in the original Project AI❤dol Public English Dataset to create phrase-level excerpts. We then convert all AP and EP to SP, and manually replace all non-ARPAbet phoneme instances with the best fitting ARBAbet replacement. This was done to make Project AIdol Segmented share the same vocabulary as our SynthGT dataset.

We share the audio files in the wav folder and the processed versions of the original Project AI❤dol Public English Dataset annotation files in the processed_annotations folder. The lab folder contains the phoneme sequence of each segment, and the TextGrid folder contains the phoneme sequence along with phoneme durations of each segment in TextGrid format. In our experiments we use the wav files along with the lab folder files to predict phoneme durations, which we compare to the ground truth TextGrid files. Our models which we have evaluated with this benchmark are available in our model collection.

If you use this dataset for your own experiments or any other purpose, please give proper attribution to the original Project AI❤dol Public English Dataset and author Lotte V (@lottev1991), and cite our work:

bibtex
@unpublished{Antonisen2026SynthGT,
  author = {Silas Antonisen and Iván López-Espejo},
  title  = {Synthesizing Ground Truth Phoneme Boundaries with SynthGT: A Synthetically Generated Solo-Singing Corpus},
  note   = {Submitted to IEEE Transactions on Audio, Speech and Language Processing},
  year   = {2026}
}