Aybee5/ha-tts-csv

Hausa TTS Dataset

Dataset Description

This dataset contains Hausa language text-to-speech (TTS) recordings from multiple speakers. It includes audio files paired with their corresponding Hausa text transcriptions.

Dataset Structure

The dataset is organized as follows:

├── data/
│   ├── metadata.csv                    # Metadata (source, audio paths, text)
│   └── audio_files/
│       ├── 97f373e8-f6e6-.../          # Speaker 1 audio files
│       ├── b0db0a87-2206-.../          # Speaker 2 audio files
│       └── c3621689-ca53-.../          # Speaker 3 audio files
└── README.md

Data Fields

The metadata.csv contains the following columns:

• file_name: Relative path to the audio file in WAV format
• source: Speaker ID (UUID format) identifying the speaker
• text: Hausa text transcription corresponding to the audio

Dataset Statistics

• Total Samples: ~100 recordings
• Number of Speakers: 3 unique speakers
• Audio Format: WAV files
• Sample Rate: 24,000 Hz (target)
• Language: Hausa (ha)

Usage

Loading the Dataset

from datasets import load_dataset

# Load from Hugging Face
dataset = load_dataset("Aybee5/ha-tts-csv", split="train")

# Access the data
print(dataset[0])
# Output: {'source': 'speaker_id', 'audio': {'path': '...', 'array': [...], 'sampling_rate': 24000}, 'text': '...'}

Example with Audio Processing

from datasets import load_dataset, Audio

# Load dataset
dataset = load_dataset("Aybee5/ha-tts-csv", split="train")

# Cast audio column to specific sampling rate
dataset = dataset.cast_column("audio", Audio(sampling_rate=24000))

# Process audio
for example in dataset:
    audio_array = example["audio"]["array"]
    sampling_rate = example["audio"]["sampling_rate"]
    text = example["text"]
    speaker_id = example["source"]
    # Your processing here...

Speaker Information

The dataset includes recordings from 3 unique speakers, each identified by a UUID in the source field. For training multi-speaker TTS models, you can map these UUIDs to numeric speaker IDs:

unique_speakers = sorted(set(dataset["source"]))
speaker_to_id = {speaker: idx for idx, speaker in enumerate(unique_speakers)}

Use Cases

This dataset is suitable for:

• Text-to-Speech (TTS) model training
• Multi-speaker voice synthesis
• Hausa language speech research
• Voice cloning applications
• Speech corpus analysis

Languages

• Hausa (ha)

Licensing Information

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

Citation

If you use this dataset in your research, please cite:

@dataset{hausa_tts_dataset,
  title={Hausa TTS Dataset},
  author={Your Name},
  year={2025},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/datasets/Aybee5/ha-tts-csv}}
}

Data Collection and Processing

The audio data was collected using MimicStudio recording system and includes native Hausa speakers reading text prompts. All audio files are stored in WAV format with metadata tracked in an SQLite database, which has been exported to CSV format for easy dataset distribution.

Considerations for Using the Data

• Audio quality may vary between speakers
• Some audio files may have background noise
• Text transcriptions are in Hausa language using Latin script
• Speaker characteristics (gender, age, accent) are not explicitly labeled

Additional Information

For questions or issues with this dataset, please open an issue on the dataset repository or contact the dataset maintainer.