Jean Kpoti

EchoVoiice Assistant is a voice assistant you talk to in Ewe, a language spoken by millions of people in Ghana and Togo that most AI tools don’t support. You speak Ewe, and it speaks Ewe back. Nobody has to type, read, or know English, so people who are shut out of today’s assistants because of language or literacy can use it. I built it as a pipeline. Meta’s MMS model turns the user’s Ewe speech into text, and Meta’s NLLB translates it when that’s needed. Muse Spark, through the Meta Model API, works out the answer. Then my EchoVoiice text-to-speech model says that answer out loud in Ewe. All of it runs in Docker behind a web page. One use I focus on is a hospital interpreter mode. It helps patients and staff understand each other at check-in and with directions, and it does not give medical advice. Next, I’m fine-tuning the models on more everyday Ewe and adding more African languages. My EchoVoiice text-to-speech API existed before this event. I built the assistant, the pipeline, and the web app during the hackathon.