| The dataset totals 84 hours, 45 minutes, and 4 seconds of data and approximately 846 speakers from multiple combined datasets such as: (VCTK + M4Singer + MIR1K + PTDB) along with several other datasets that, frankly, I no longer remember exactly what they were. | |
| Feature files were extracted using RMVPE and Contentvec base. | |
| Speaker 435 may cause errors during training due to omissions during dataset assembly. This can be temporarily resolved by adjusting sid and spk_embed_dim to 847. | |
| Dataset | |
| - Total raw dataset size is approximately 28 GB. | |
| - After extraction, the dataset occupies around 117 GB (excluding spec files). | |
| - Spec files take up an additional approximately 125 GB. | |
| - Total processed dataset size: 117 GB + 125 GB = 242 GB | |
| WARNING: THIS DATASET IS FOR MODEL EXPERIMENT ONLY AND ITS QUALITY IS NOT 100% GUARNTEED. |