README.md

---
license: cc-by-4.0
language:
- sw
pretty_name: Afrivoice ASR Swahili dataset
task_categories:
- automatic-speech-recognition
tags:
- DigitalUmuganda
- DU
- sw
- swa
- Swahili
- asr
- stt
- voice
- speech
size_categories:
- 100K<n<1M
annotations_creators:
- crowdsourced
language_creators:
- crowdsourced
---
# Dataset Card for the image text and voice dataset


## Dataset Description

This is an image prompt ASR dataset for Swahili. The dataset was collected on 5 domains: Agriculture, Education, Finance, Government and Health.

<!-- | Domain       | Total Hours | Transcribed Hours | Number of Clips | Dataset Size (GB) |
|--------------|-------------|-------------------|-----------------|-------------------|
| Agriculture  | 713.3       | 692.11            | 124,648         | 24.208            |
| Education    | 408.43      | 392.33            | 70,745          | 14.439            |
| Finance      | 611.66      | 601.15            | 105,763         | 24.127            |
| Government   | 578.14      | 566.15            | 101,021         | 23.921            |
| Health       | 668.29      | 608.33            | 118,827         | 22.438            |
| **Total**    | **2979.81** | **2860.07**       | **521,004**     | **109.133**       | -->


| Domain      |   Total Hours |   Transcribed Hours |   Number of Clips |   Dataset Size (GB) |
|:------------|--------------:|--------------------:|------------------:|--------------------:|
| Agriculture |        739.67 |              717.66 |            128578 |             43.537  |
| Education   |        414.3  |              398.1  |             71823 |             30.5219 |
| Financial   |        615.29 |              604.78 |            106387 |             57.8098 |
| Government  |        591.1  |              579.11 |            102915 |             49.6434 |
| Health      |        677.92 |              617.96 |            120543 |             37.3837 |
| **Total**   |   **3038.28** |         **2917.61** |        **530246** |        **218.896**  |

## 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 by using the `load_dataset` function. 

### For full dataset

```python
from datasets import load_dataset
data = load_dataset("DigitalUmuganda/Afrivoice_Swahili")
```

### For categorical dataset

```python
from datasets import load_dataset
data = load_dataset("DigitalUmuganda/Afrivoice_Swahili", name="health")
```

The categories options are:
- agriculture
- education
- financial
- government
- health

## Dataset Structure

### Data Instance



```python
{
  "voice_creator_id":"3bqVVjMdPtUg3YDJE2T0b8jUuTq1",
  "transcription_creator_id":"GZbQSWNUOoZDNgTSKlz7ZZgXAHC2",
  "image_filepath":"32749ce.jpg",
  "image_category":"agriculture",
  "image_sub_category":"food banks",
  "category":"agriculture",
  "audio_filepath":"xadnlONijhyVvN2USCkF.webm",
  "transcription":"Kuna wamama watatu ambao wanaonekana karibu na magunia ambayo ni ya rangi nyeupe, na pia kuna [cs]katoni[cs], karibu nao kuna mlima ambao unaonekana pamoja na mahindi, na juu pia kuna mawingu ya rangi nyeupe, na ya rangi ya samawati.",
  "normalized_transcription":"kuna wamama watatu ambao wanaonekana karibu na magunia ambayo ni ya rangi nyeupe na pia kuna cskatonics karibu nao kuna mlima ambao unaonekana pamoja na mahindi na juu pia kuna mawingu ya rangi nyeupe na ya rangi ya samawati",
  "age_group":"25-35",
  "gender":"Male",
  "project_name":"Digital Umuganda",
  "locale":"sw_KE",
  "year":2025,
  "duration":18.3,
  "location":null,
  "uid":"3bqVVjMdPtUg3YDJE2T0b8jUuTq1",
  "key":"xadnlONijhyVvN2USCkF",
  "dir_path":"agriculture_swahili_dev",
  "chunk_id":0
}



```

### Data Fields

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

`transcription_creator_id` (`string`): An id for which client (text) made the transcription

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

`normalized_transcription` (`string`): Original audio transcription without punctuation and capitalization

`image_filepath` (`string`): file path of the image file inside the dataset

`audio_filepath` (`string`): file path of the audio file inside the dataset

`dir_path` (`string`): main directory path of the record

`chunk_id` (`string`): chunk identification number where the record belong

`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.

`category` (`string`): category of the record

`project` (`string`): project name

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

`year` (`int`): Year of recording

`project_name` (`string`): project name

`location` (`string`): location of the project

`key` (`string`): key identifier of the data point

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