The training dataset only contains synthesized data. Is there any labeled data available? Also, does the author have a processed dataset with labels that could be provided?

[DommyWorld]

[JusperLee]

Thanks for your question! Regarding the labeled data: Yes, there is labeled metadata available for each sequence in the dataset. The metadata is stored in each folder alongside the synthesized data files.

https://huggingface.co/datasets/JusperLee/SonicSet-v2

[DommyWorld]

Thanks for your question! Regarding the labeled data: Yes, there is labeled metadata available for each sequence in the dataset. The metadata is stored in each folder alongside the synthesized data files.

https://huggingface.co/datasets/JusperLee/SonicSet-v2

Thank you for your reply. I have checked the link you provided and downloaded the dataset. Perhaps my previous explanation was not clear. What I meant is that during the data synthesis process, you used the original LibriSpeech data by segmenting and concatenating it before performing simulation. What I want to ask is whether there are any speech audio files from this segmented and concatenated process that have not undergone simulation.