| --- |
| pretty_name: Bagpiper Pretraining Data Collection |
| language: |
| - en |
| task_categories: |
| - automatic-speech-recognition |
| - text-to-audio |
| tags: |
| - audio |
| - speech |
| - music |
| - sound |
| - multimodal |
| - rich-caption |
| - opus |
| size_categories: |
| - 100M<n<1B |
| --- |
| |
| # Bagpiper Pretraining Data Collection |
|
|
| **Upload in progress. This repository is not yet complete and should not be |
| used or cited until this notice is removed.** |
|
|
| Opus-compressed audio with rich captions and curation subset labels, in the |
| archive layout the ESPnet speechlm dataloader reads directly. |
|
|
| This is the compressed, directly trainable counterpart to |
| [espnet/Bagpiper_PreTrain_Data](https://huggingface.co/datasets/espnet/Bagpiper_PreTrain_Data), |
| which ships HuggingFace-style Parquet that the training code cannot load. |
|
|
| Planned contents: 16 kHz mono Ogg/Opus at about 32 kbps, roughly 20 TB, with |
| each utterance carrying its rich caption as a plain string and its membership |
| in the curation subsets (`speech_und`, `speech_gen`, `sound_und`, `sound_gen`, |
| `music_und`, `music_gen`, plus the SFT-sized variants). An utterance belongs to |
| zero, one or two of the base subsets. |
|
|
| A full dataset card, per-dataset statistics and usage instructions replace this |
| notice when the upload completes. |
|
|
