KUMA

KUMA is an API designed for voice intent detection in the Mina and Ewe languages, primarily spoken in West Africa. It addresses the challenge faced by millions of Mina and Ewe speakers whose languages are not understood by common ride-hailing, delivery, and service applications. By enabling these apps to process spoken commands in Mina and Ewe, KUMA facilitates more accessible and inclusive digital services for these linguistic communities.

The platform collects audio recordings in Mina and Ewe to train a model capable of interpreting user intent from spoken phrases. When an audio file is submitted to the API, it returns a structured intent, such as requesting a ride or a delivery, along with a confidence score, extracted entities like destination, and a transcription of the spoken command. Supported audio formats include OGG, WAV, MP3, M4A, and WebM, with a maximum file size of 25 MB. The detection process is asynchronous: after submitting an audio file via POST, the API immediately provides a request ID, and the caller retrieves the detected intent by polling with that ID.

KUMA is intended for developers and organizations looking to integrate Mina and Ewe voice intent detection into their applications. Access to the API requires creating an account and submitting a use case, after which credentials are provided if approved. Authentication uses a JWT-based system, where the client secret is exchanged for a token valid for one hour, enhancing security by limiting credential exposure.

Each account receives 100 free API calls per month. If the quota is exceeded, users are instructed to contact the service for further access. The platform is based in Lomé, Togo, and is currently available in a beta version. KUMA positions itself as a voice intent detection API for underrepresented languages, supporting integration with a variety of apps that serve Mina and Ewe speakers.