---
language:
- fon
license: cc-by-4.0
task_categories:
- automatic-speech-recognition
pretty_name: Fongbe ASR
size_categories:
- 1K<n<10K
tags:
- fongbe
- asr
- speech
- african-languages
- low-resource
dataset_info:
  features:
  - name: audio
    dtype: audio
  - name: sentence
    dtype: string
  - name: speaker
    dtype: string
  - name: duration
    dtype: float64
  splits:
  - name: train
    num_examples: 8234
  - name: test
    num_examples: 2168
---

# Fongbe ASR Dataset

Speech recognition dataset for **Fongbe** (fɔ̀ngbè), a Gbe language spoken by approximately 4.1 million people, primarily in Benin and Togo.

## Dataset Description

This dataset contains read speech recordings in Fongbe with their transcriptions. It was originally collected by Fréjus A. A. Laleye and organized in the [pyFongbe](https://github.com/laleye/pyFongbe) repository.

### Statistics

| Split | Samples | Duration | Speakers |
|-------|---------|----------|----------|
| train | 8,234   | ~5.7h    | 23       |
| test  | 2,168   | ~1.5h    | 4        |
| **total** | **10,402** | **~7.2h** | **27** |

### Audio Format

- Format: WAV (PCM 16-bit)
- Sample rate: 16,000 Hz
- Channels: mono

## Usage

```python
from datasets import load_dataset

ds = load_dataset("alaleye/fon")

# Access a sample
sample = ds["train"][0]
print(sample["sentence"])   # Fongbe transcription
print(sample["audio"])      # {'array': np.array(...), 'sampling_rate': 16000}
print(sample["speaker"])    # Speaker name
print(sample["duration"])   # Duration in seconds
```

### Streaming

```python
ds = load_dataset("alaleye/fon", streaming=True)
for sample in ds["train"]:
    print(sample["sentence"])
    break
```

## Features

- `audio` (`Audio`): audio waveform sampled at 16kHz
- `sentence` (`string`): Fongbe transcription
- `speaker` (`string`): speaker identifier
- `duration` (`float64`): audio duration in seconds

## Speakers

**Train set (23 speakers):** armandine, boris, davacan, denise, dolores, donald, emmanuel, eunice, hans, herman, inconnu, lorseque, mario, melissa, miguel, mikael, narcisse, nazer, osias, parisius, rose, sorel, stephanie

**Test set (4 speakers):** antoine, cyrielle, frejus, helmut

Train and test speakers are disjoint, ensuring speaker-independent evaluation.

## Source

- **Repository:** [laleye/pyFongbe](https://github.com/laleye/pyFongbe)
- **Related project:** [ALFFA (African Languages in the Field: speech Fundamentals and Automation)](http://www.openslr.org/25/)

## Citation

If you use this dataset, please cite:

```bibtex
@inproceedings{laleye:hal-01436788,
  TITLE = {{First Automatic Fongbe Continuous Speech Recognition System: Development of Acoustic Models and Language Models}},
  AUTHOR = {Laleye, Fr{\'e}jus a A and Besacier, Laurent and Ezin, Eug{\`e}ne C and Motamed, Cina},
  URL = {https://hal.science/hal-01436788},
  BOOKTITLE = {{Proceedings of the Federated Conference on Computer Science and Information Systems}},
  ADDRESS = {Gdansk, Poland},
  VOLUME = {8},
  PAGES = {477 - 482},
  YEAR = {2016},
  MONTH = Sep,
  DOI = {10.15439/2016F153},
  KEYWORDS = { corpus ;  African languages ;  Fongbe ;  under-resourced languages ; automatic speech recognition (ASR)},
  PDF = {https://hal.science/hal-01436788v1/file/153.pdf},
  HAL_ID = {hal-01436788},
  HAL_VERSION = {v1},
}
```