Wolof Speech Synthesis Recordings from Two Native Speakers

Baamtu Datamation created this Wolof Text-to-Speech dataset under the AI4D African language program. It contains recordings from a male speaker for 22 hours 28 minutes 41 seconds and a female speaker for 18 hours 47 minutes 19 seconds, each contributing over 20,000 sentences sourced from news and Wikipedia.

Use Cases

• Train a speech synthesis model on the male and female speaker audio recordings to generate Wolof speech.
• Fine-tune a TTS system using the over 20,000 sentences per speaker to improve naturalness and prosody.
• Analyze phonetic and prosodic patterns in Wolof by comparing the 22-hour male and 18-hour female audio corpora.
• Develop voice cloning models for Wolof by leveraging the distinct male and female speaker recordings.

Strengths

• Substantial audio duration totaling over 41 hours from two speakers.
• Clear speaker diversity with separate male and female voice recordings.
• Large text corpus with over 20,000 unique sentences per speaker.

Limitations

• No column or metadata schema is provided, complicating structured access.
• Sample data and file formats are unspecified, hindering immediate usability.
• Potential bias towards formal text domains like news and Wikipedia.

Provenance

Source

Baamtu Datamation, hosted on Zenodo.

Collection Method

Recordings collected from two native Wolof speakers.

Freshness

Last updated on 2024-08-26.

Geography

Senegal (implied by language and tags).