Pidgin ASR Combined: Nigerian Pidgin Speech-to-Text Dataset with 8.6 Hours of Audio

Pidgin ASR Combined is a unified Nigerian Pidgin English speech-to-text dataset created by michaelodafe. It contains approximately 8.6 hours of audio across 4,278 clips from 10 source speakers, formatted as 16 kHz mono WAV files. The dataset was last updated on 2026-05-13 and was used to train a Whisper model that achieved a 21.37% word error rate.

Use Cases

• Fine-tuning automatic speech recognition models based on the described Nigerian Pidgin audio clips.
• Benchmarking model performance against the reported 21.37% word error rate on the test split.
• Training or evaluating models for low-resource language speech recognition based on the combined Pidgin sources.

Strengths

• Consolidates multiple public Pidgin ASR sources into a single, consistent schema with defined train/validation/test splits.
• Contains approximately 8.6 hours of audio data across 4,278 clips, providing a foundation for model training.
• The derived Whisper model achieved a 21.37% word error rate, demonstrating the dataset's utility for model development.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset size in terms of rows is unknown, which may limit suitability assessment.
• Last updated 2026-05-13 10:18:51; freshness should be verified.

Provenance

Source

huggingface, author michaelodafe

Collection Method

Combination of publicly available Nigerian Pidgin ASR sources.

Freshness

2026-05-13

Geography

Nigeria (implied by focus on Nigerian Pidgin)