| --- |
| configs: |
| - config_name: default |
| data_files: |
| - split: train |
| path: data/train/*.parquet |
| - split: validation |
| path: data/validation/*.parquet |
| --- |
| |
| <!-- # Taiwan TTS Corpus for VoxCPM2 |
|
|
| Prepared from the SoundOn Taiwan TTS build for VoxCPM2 fine-tuning. |
|
|
| ## Contents |
|
|
| - `data/train/*.parquet` and `data/validation/*.parquet`: Hugging Face parquet shards. |
| - `audio` column is written with `datasets.Audio(sampling_rate=16000)` via `cast_column`. |
| - `manifests/train.jsonl`, `manifests/val.jsonl`, and `manifests/metadata.jsonl` are included for reproducibility. |
|
|
| ## Build Summary |
|
|
| - Records: 17078236 |
| - Train: 17060993 |
| - Validation: 17243 |
| - Hours: 47825.278 |
| - Speakers: 2500 |
| - Podcasts: 703 |
|
|
| ## Rights |
|
|
| Current status: `private_training_pending_rights_review`. |
| Do not make the dataset public until redistribution and speaker/voice consent are confirmed. |
| --> |
