|
|
--- |
|
|
pretty_name: Bobi Wine Luganda Speech |
|
|
language: |
|
|
- lg |
|
|
license: unknown |
|
|
task_categories: |
|
|
- text-to-speech |
|
|
tags: |
|
|
- luganda |
|
|
- uganda |
|
|
- speech |
|
|
- tts |
|
|
size_categories: |
|
|
- 1K<n<10K |
|
|
configs: |
|
|
- config_name: default |
|
|
data_files: |
|
|
- split: train |
|
|
path: data/train-* |
|
|
dataset_info: |
|
|
features: |
|
|
- name: audio |
|
|
dtype: audio |
|
|
- name: transcription |
|
|
dtype: string |
|
|
splits: |
|
|
- name: train |
|
|
num_bytes: 116640417 |
|
|
num_examples: 250 |
|
|
download_size: 116308260 |
|
|
dataset_size: 116640417 |
|
|
--- |
|
|
|
|
|
# Bobi Wine Luganda Speech |
|
|
|
|
|
This dataset contains short Luganda speech clips segmented from longer recordings, with corresponding transcriptions in Luganda. |
|
|
It is intended for Text-to-Speech (TTS) research and prototyping. |
|
|
|
|
|
## Contents |
|
|
- `wavs/`: 250 WAV clips (`22050 Hz` recommended for TTS pipelines) |
|
|
- `metadata.csv`: Root-level metadata with two columns per line: `path|text` |
|
|
- Example: `wavs/0001.wav|Gyebale ko ssebo…` |
|
|
- `bobi_interview_clean.wav`: Source interview audio (cleaned) |
|
|
- `bobi.mp3`: Additional source audio |
|
|
|
|
|
## Recommended metadata format for TTS |
|
|
Most TTS pipelines expect a root-level `metadata.csv` with relative paths: |
|
|
|
|
|
``` |
|
|
path|text |
|
|
wavs/0001.wav|Gyebale ko ssebo… |
|
|
wavs/0002.wav|… |
|
|
``` |
|
|
|
|
|
If you need to regenerate `metadata.csv` from an ID|text file: |
|
|
|
|
|
```bash |
|
|
cd /Users/jovan/dev/ai_slop/dataset/bobiwine |
|
|
awk -F'|' 'NF>=2 { gsub(/^[ \t]+|[ \t]+$/, "", $1); print "wavs/" $1 ".wav|" $2 }' chops/metadata.csv > metadata.csv |
|
|
``` |
|
|
|
|
|
Note: Ensure that transcriptions are Luganda and match the audio content. |
|
|
|
|
|
## Upload to Hugging Face Hub |
|
|
|
|
|
```bash |
|
|
curl -LsSf https://hf.co/cli/install.sh | bash |
|
|
hf auth login |
|
|
|
|
|
cd /Users/jovan/dev/ai_slop/dataset/bobiwine |
|
|
# Ensure large audio files are tracked with LFS via .gitattributes |
|
|
|
|
|
# Option A: Use the HF CLI upload (simple) |
|
|
hf upload unreal/bobiwine . --repo-type=dataset |
|
|
|
|
|
# Option B: Git-based workflow |
|
|
# git init && git remote add origin https://huggingface.co/datasets/unreal/bobiwine |
|
|
# git add . |
|
|
# git commit -m "Add Bobi Wine Luganda speech dataset" |
|
|
# git push -u origin main |
|
|
``` |
|
|
|
|
|
## Caveats |
|
|
- Licensing for the audio should be reviewed and set appropriately. |
|
|
- Content may contain political speech; review for intended downstream use. |
|
|
|
|
|
## Citation |
|
|
If you use this dataset, please cite and credit the dataset author and original sources as appropriate. |
|
|
|
|
|
|
|
|
|
