Multilingual Nigerian Speech Corpus with 1,800 Hours

1,800 hours of authentic speech data across three major Nigerian languages: Igbo, Hausa, and Yoruba. The dataset contains audio from over 5,000 diverse speakers, with approximately 600 hours per language, and was created by the NaijaVoices project. It was last updated on the platform in August 2025.

Use Cases

• Train an ASR model to transcribe Yoruba audio files into text transcripts.
• Fine-tune a multilingual speech model using Igbo and Hausa speech segments with corresponding curated text.
• Analyze speaker demographics and speech patterns using metadata on age and gender distribution.
• Develop voice synthesis systems for Nigerian languages using the high-quality, speaker-diverse audio samples.

Strengths

• 1,800 total hours of speech data provides substantial volume for model training.
• Over 5,000 unique speakers ensures significant speaker diversity.
• Balanced collection with ~600 hours each for Igbo, Hausa, and Yoruba languages.
• Curated text annotations and balanced age and gender representation enhance utility.

Limitations

• Specific row count, column names, and file formats are not provided in the input.
• The dataset's geographic coverage is focused on Nigeria, limiting generalizability to other regions.
• License terms are not specified, which could restrict commercial use.

Provenance

Source

NaijaVoices project (naijavoices).

Collection Method

Collected from authentic speech recordings of over 5,000 diverse speakers, with expert-curated text.

Freshness

Last updated on the platform in August 2025.

Geography

Nigeria, covering speakers of Igbo, Hausa, and Yoruba languages.