The area of Automatic Speech Recognition (ASR) is of particular importance, as it is crucial to provide voice services (recognition and speaking) to underserved groups. However, it remains unclear how much data needs to be collected to achieve a “good” Automatic Speech Recognition (ASR) model in low-resource languages. It has been found that the amount of speech data required for ASR model training data has a significant influence on the robustness of ASR systems. Moreover, it is also unclear how much data is required to build domain-specific ASR models. The objective of the research is to develop an evidence base for the amount of speech data required to build a good automatic speech recognition model across priority “low-resource” African languages