Ghana Speech: Multilingual Audio Dataset for Low-Resource African Languages

Ghana Speech is a multilingual audio dataset for speech recognition, shared by the ghananlpcommunity under a CC BY-NC 4.0 license. The dataset includes speech segments in several Ghanaian languages, such as Akuapem Twi with 52,650 segments (63.25 hours) and Asante Twi with 143,383 segments (200.02 hours). It was last updated on 2026-07-21.

Use Cases

• Train automatic speech recognition models based on the multilingual audio segments.
• Benchmark speech synthesis systems for languages like Akuapem Twi and Asante Twi.
• Conduct linguistic research on the phonetics and structure of Ghanaian languages using the audio data.
• Develop language identification models for African languages based on the provided language subsets.

Strengths

• Includes substantial audio duration for major languages, such as 200.02 hours for Asante Twi.
• Covers multiple low-resource Ghanaian languages, including Akuapem Twi, Anyin, and Avatime.
• Provides segment counts per language, such as 143,383 segments for Asante Twi.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The license is CC BY-NC 4.0, restricting commercial use.

Provenance

Source

ghananlpcommunity

Freshness

Last updated 2026-07-21 12:29:06; freshness should be verified.

Geography

Likely Ghana, based on the languages included.