README.md

---
configs:
- config_name: agriculture
  data_files:
  - split: train
    path: agriculture_train_tarred/sharded_manifests_with_image/manifest_*.json
  - split: validation
    path: agriculture_val_tarred/sharded_manifests_with_image/manifest_*.json
  - split: test
    path: agriculture_test_tarred/sharded_manifests_with_image/manifest_*.json
- config_name: education
  data_files:
  - split: train
    path: education_train_tarred/sharded_manifests_with_image/manifest_*.json
  - split: validation
    path: education_val_tarred/sharded_manifests_with_image/manifest_*.json
  - split: test
    path: education_test_tarred/sharded_manifests_with_image/manifest_*.json
- config_name: financial
  data_files:
  - split: train
    path: financial_train_tarred/sharded_manifests_with_image/manifest_*.json
  - split: validation
    path: financial_val_tarred/sharded_manifests_with_image/manifest_*.json
  - split: test
    path: financial_test_tarred/sharded_manifests_with_image/manifest_*.json
- config_name: government
  data_files:
  - split: train
    path: goverment_train_tarred/sharded_manifests_with_image/manifest_*.json
  - split: validation
    path: goverment_val_tarred/sharded_manifests_with_image/manifest_*.json
  - split: test
    path: goverment_test_tarred/sharded_manifests_with_image/manifest_*.json
- config_name: health
  data_files:
  - split: train
    path: health_train_tarred/sharded_manifests_with_image/manifest_*.json
  - split: validation
    path: health_val_tarred/sharded_manifests_with_image/manifest_*.json
  - split: test
    path: health_test_tarred/sharded_manifests_with_image/manifest_*.json
default: agriculture
license: cc-by-4.0
task_categories:
- automatic-speech-recognition
language:
- rw
tags:
- ASR
- STT
- RW
- KIN
pretty_name: Afrivoice Kinyarwanda ASR dataset
size_categories:
- 100K<n<1M
annotations_creators:
- crowdsourced
language_creators:
- crowdsourced
---

# Dataset Card for the image text and voice dataset


## Dataset Description

Each datapoint in this dataset consists of a JPEG image, a corresponding audio Webm file describing the image, and when available, the transcription of the audio file. 


| Domain      | Total Hours | Transcribed Hours | Number of Clips | Dataset Size (GB) |
|-------------|-------------|-------------------|-----------------|-------------------|
| Agriculture | 467.13      | 465.40            | 86,305          | 30.13             |
| Health      | 994.32      | 992.87            | 179,219         | 58.53             |
| Finance     | 564.21      | 563.11            | 103,159         | 38.55             |
| Government  | 676.10      | 674.11            | 122,265         | 49.22             |
| Education   | 198.49      | 197.52            | 37,377          | 13.70             |
| Scripted Education | 262.63   | 262.63        |    42,062       | 15.08             |
| **Total**   | **3162.88** | **3155.64**       | **570,387**     | **205.21**        |


## How to use

The `datasets` library allows you to load and pre-process your dataset in pure Python, at scale. The dataset can be downloaded and prepared in one call to your local drive in two ways: 

Direct download of the whole dataset to a local directory
```
from huggingface_hub import snapshot_download                                                   

snapshot_download(repo_id="DigitalUmuganda/Afrivoice_Kinyarwanda",repo_type='dataset',local_dir='<destination_dir>')
```
Or, using load_dataset to download a particular domain
```
from datasets import load_dataset

data = load_dataset("DigitalUmuganda/Afrivoice_Kinyarwanda",name='health')
```


## Dataset Structure

### Data Instances


```python
{'creator_id': 'bcSXMYbErjM6pJyAwwLs7NAxA9v2',
 'raw_text': 'Ingagi ihagaze yonyine. Ingagi ni nziza cyane, kuko zikurura ba mukerarugendo bakazana amadovize mu Gihugu cyacu.',
 'duration': 15.06,
 'LUFS': -25.5,
 'image_category': 'Agriculture',
 'image_sub_category': 'Wild Animals',
 'text': 'ingagi ihagaze yonyine ingagi ni nziza cyane kuko zikurura ba mukerarugendo bakazana amadovize mu gihugu cyacu',
 'audio_filepath': 'audio_1751479904-bcSXMYbErjM6pJyAwwLs7NAxA9v2.webm',
 'image_filepath': 'restyf.jpg',
 'age_group': '50+',
 'gender': 'Male',
 'location': 'Musanze',
 'shard_id': 0,
 'image_shard_id': 0}


```

### Data Fields

`creator` (`string`): An id for which client (voice) made the recording

`raw_text` (`string`): Original audio transcription with punctuation and capitalization

`image_filepath` (`string`): name of the image file inside the shard

`audio_filepath` (`string`): name of the audio file inside the shard

`text` (`string`): normalized audio transcription (i.e: without punctuation and capitalization)

`age_group` (`string`): age range of the audio recorder 

`gender` (`string`): The gender of the speaker

`location` (`string`): geographical location of the audio recorder

`duration` (`int`): length in seconds of the audio file

`image_category` (`string`): domain of the image (eg: health, agriculture, finance), used as prompt during audio creation.

`image_sub_category` (`string`): Sub-domain label of the image (e.g., within agriculture: “seed farming” or “forestry”), used to guide audio creation.

`shard_id` (`int`): index of the shard containing the audio file in the `audio_filepath` column.

`image_shard_id` (`int`): index of the shard containing the image in the `image_filepath` column.

`LUFS` (`int`): Loudness of the audio in Loudness Units relative to Full Scale (LUFS). Lower values indicate quieter audio, while higher values indicate louder audio.

### Data Splits

Each domain in the dataset is divided into train, validation, and test splits.

# Licensing Information
All datasets are licensed under the [Creative Commons license (CC-BY-4)](https://creativecommons.org/licenses/).