This paper considers a dimension of the colonial archive that records, often in unsettling detail, white views on colonial, post-independence, and apartheid subjects in Africa during the latter half of the twentieth century: the sound archive of the Africa Evangelistic Band (AEB). Founded in Cape Town in 1924 by three Irish sisters, the AEB combined evangelical zeal with the infrastructures of colonial modernity, expanding its missionary work from the Cape Peninsula to what was then Bechuanaland, South West Africa, and the Belgian Congo. Its archive - over 6,000 cassette tapes and 78 rpm recordings of sermons, testimonies, and songs in Afrikaans, Xhosa, Setswana, and English - offers a rare record of multilingual religious encounters embedded in colonial and apartheid contexts.
In this presentation, the speaker focuses on the experimental use of automatic speech recognition (ASR) and large language models to transcribe and analyze these recordings. While ASR performs adequately in English and Afrikaans, it fails almost entirely in isiXhosa, exposing the algorithmic bias that continues to marginalize African languages in digital infrastructures.
The speaker proposes a “good enough AI” approach - combining imperfect automation, orthographic normalization, and human oversight - to open the archive pragmatically, while foregrounding its affective, theological, and racial resonances.
