Afrispeech Parliament

AfriSpeech-Parliament is a 35.86-hour collection of transcribed parliamentary sessions from Nigeria, Ghana, South Africa, and Kenya. The dataset, created by intronhealth, consists of 16-second audio chunks sourced from official YouTube channels, each paired with manual transcriptions. It was last updated on the Hugging Face platform on April 11, -2025.

Use Cases

• Train automatic speech recognition (ASR) models based on parliamentary audio and manual transcriptions.
• Analyze linguistic patterns and political discourse based on transcribed parliamentary speech.
• Benchmark model performance on African accents and languages based on the four-nation geographic coverage.
• Develop tools for parliamentary speech analysis and summarization based on the structured audio-text pairs.

Strengths

• Contains 35.86 hours of manually transcribed audio data.
• Sourced from official parliamentary YouTube channels for four African nations.
• Provides structured audio-text pairs with 16-second audio chunks.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect geographic and source bias inherent to the selected four nations and YouTube channels.

Provenance

Source

Official parliamentary YouTube channels of Nigeria, Ghana, South Africa, and Kenya.

Collection Method

Audio chunks extracted and manually transcribed.

Time Range

null

Freshness

Last updated 2025-04-11 17:07:29; freshness should be verified.

Geography

Nigeria, Ghana, South Africa, Kenya