Rexe/nigerian-pidgin-speech

Nigerian Pidgin Audio + Text Dataset for Whisper Fine-tuning

Nigerian Pidgin speech dataset for Whisper fine-tuning

Dataset Summary

This dataset contains audio recordings and transcriptions in Nigerian Pidgin English, designed for fine-tuning speech recognition models, particularly OpenAI's Whisper.

Dataset Structure

• Train Split: 65 samples
• Test Split: 8 samples
• Total Duration: 0.0 hours (estimated)
• Average Duration: 2.4 seconds per sample
• Sample Rate: 16kHz
• Audio Format: WAV

Sample Data

Here are some example transcriptions from the dataset:

- ♪ EVERYDAY NA BILLING ♪
- ♪ SHE CALL ME SANTA ♪
- ♪ SEYCHELLES, ADDIS ABABA ♪
- ♪ I FOR DON TURN VISITOR ♪
- ♪ SHE NO WAN TURN TO OROBOKIBO ♪

Usage

Loading the Dataset

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("Rexe/nigerian-pidgin-speech")

# Access train and test splits
train_data = dataset["train"]
test_data = dataset["test"]

Training with Transformers

from transformers import WhisperForConditionalGeneration, WhisperProcessor

# Load model and processor
model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-small")
processor = WhisperProcessor.from_pretrained("openai/whisper-small")

# Your training code here...

Google Colab Training

Use this dataset directly in Google Colab for training:

# Install requirements
!pip install datasets transformers torch torchaudio

# Load dataset
from datasets import load_dataset
dataset = load_dataset("Rexe/nigerian-pidgin-speech")

# Start training...

Languages

• English (en): Base language
• Nigerian Pidgin (pcm): Target language for fine-tuning

Common Pidgin Phrases

• "How you dey?" - How are you?
• "I dey fine o" - I am fine
• "Wetin dey happen?" - What's happening?
• "Make we go" - Let's go
• "Abeg help me" - Please help me

Citation

If you use this dataset, please cite:

@dataset{nigerian_pidgin_speech,
  title={Nigerian Pidgin Audio + Text Dataset},
  author={Your Name},
  year={2024},
  url={https://huggingface.co/datasets/Rexe/nigerian-pidgin-speech}
}

License

This dataset is released under the MIT License.