# Automatic Speech Recognition in Wolof — Data

来源：https://www.kaggle.com/competitions/asr-in-wolof/data

采集时间：2026-10-20T09:04:59Z

## Dataset

Participants in this competition will work with a Wolof Automatic Speech Recognition (ASR) dataset, hosted on Hugging Face. This dataset is designed to enable the development of ASR models capable of transcribing Wolof speech, a valuable step in creating accessible technology for this low-resource language. The dataset is available in two parts:

- **[Training Data](https://huggingface.co/datasets/IndabaxSenegal/asr-wolof-dataset)**: Contains labeled audio samples for model training and validation.
- **[Test Data](https://huggingface.co/datasets/IndabaxSenegal/asr-wolof-dataset-test)**: Contains audio samples for evaluation without transcriptions, as participants will be predicting the transcriptions for this set.

### Dataset Structure

The dataset includes 13,000 samples in the **training set** and 2,000 samples in the **test set**. Each set contains essential metadata, allowing participants to work directly with audio paths and associated information.

### Columns

Both the training and test sets share the following columns, with one exception in the test set as described:

- **text**: The transcription of the audio sample. **This column is only available in the training set.** The test set does not include transcriptions, as these are the targets for model predictions.
- **duration**: The length of each audio file, measured in seconds.
- **file_name**: A unique identifier for each audio file, aiding in data organization and reference.
- **path**: The relative file path to access each audio sample.
- **audio**: Contains the raw audio data, which can be used directly by most machine learning frameworks or preprocessed according to your model’s requirements.

### Prediction Task

Participants will use the test set to generate **text transcriptions** for the provided audio files, and submissions will be evaluated based on **Word Error Rate (WER)**, a metric that measures transcription accuracy.

This structure allows for efficient training, testing, and evaluation, empowering participants to build effective ASR solutions for Wolof.