Technology that relies on data influences many facets of our lives all over the world. As this form of technology uptake increases worldwide, however, caution must be taken not to equate such ubiquity with increased utility. This is due to the fact that some of these technologies do not entirely integrate with the communities within which they are deployed in, at least in Africa. This is particularly visible within the space of voice enabled applications, most of which place heavy emphasis on widely spoken languages such as English with minimal applications offering the option of African languages. However, this is changing as the need for technologies that are responsive to the specificities of different groups becomes greater.
Voice technologies in Africa
This need for culturally grounded voice applications emerges from the existence of diverse ethnic communities in the continent coupled with relatively high uptake of information and communication technologies. Whilst the need exists, the process of filling in the gap requires dedication of skills and resources not extensively available to many people or entities on the continent. Having recognised the general lack of diverse datasets to train machine learning models especially for low resource languages, Mozilla through its Common Voice platform has created voice datasets of locally spoken languages such as Swahili and Kinyrwanda. Leveraging on the power of communities to collect these data, Mozilla hosts open voice datasets on its Common Voice platform that can be used to feed speech recognition software.
Several factors make voice technologies poised for success in Africa hence the need for such customization. Before the advent of information and communication technologies, oral traditions made up an important part of the history of many communities in Africa. The dominance of oral traditions was interrupted as the need for maintaining written records and archival grew. If any analogy is to be made then, voice technologies are a continuation of a tradition that was well anchored in most African communities. Beyond this, other factors include low digital literacy levels in a context of also low formal literacy levels that compound the challenge of having to interact with technology that requires written prompts. Additionally, the increased uptake of mobile phones, especially basic feature phones that may not have the capacity to collect any other biometric data other than voice data places speech to text technologies at a prime point for uptake.
The Privacy Concern
Up to now advancements in technology have attracted the attention of human rights advocates due to the close nexus between technology development and the impact on human rights. For voice technologies, the right to privacy has been one such right. The imperative to protect the right to privacy in voice technologies emanates from the fact that voice data is categorized as biometric data that can uniquely identify a person under the data protection laws of various countries. More specifically, it is viewed as sensitive personal information.
As a form of sensitive personal data, this implies stronger safeguards voice datasets. Each voice has a distinctive pattern as individual and identifiable as a fingerprint. Unlike other forms of personal data, voiceprints are highly unique and unchangeable hence the need for high levels of protection. This reality becomes more pertinent especially where applications built on such datasets will collect more voice data from users and aim at serving marginalized communities such as in rural areas. These scenarios will necessitate that developers be cognisant of foundational concepts and frameworks of privacy provided under data protection laws such as privacy by design to inform their design decisions when creating voice enabled applications. Equipping themselves with such practical privacy enhancing tools will reduce the disproportionate burden imposed by measures such as providing consent on marginalized communities that already face numerous instances of exclusion.