Introducing Soga Preview

Try Soga swa-csm-1b on Hugging Face

At Nadhari AI, our mission is to advance frontier AI research and applications in Sub-Saharan Africa. We believe in this mission truly. We spend a lot of time thinking about how we can build an abundant future and useful applications of AI that amplify human agency and intelligence augmentation.

One of the key modalities to achieve this goal is audio.

Today, we are introducing Soga, our Swahili voice AI app with two incredible personas: Asha and Mosi.

Powered by swa-csm-1b

Soga is powered by swa-csm-1b, the best open-source Swahili text-to-speech model. We fine-tuned Sesame's CSM-1B and achieved state-of-the-art performance in Swahili text-to-speech. The model nails pronunciation across various Swahili accents and dialects.

swa-csm-1b is now freely available on Hugging Face for developers to try out.

Voice Samples

Here are some samples showcasing the prosody of our model across different speakers, accents, and styles:

"Changanya unga na maji, hadi upatikane mchanganyiko laini."

"Hivi sasa ni saa tatu asubuhi."

"Jua linawaka sana leo, tutaenda kuogelea baharini."

"Habari za asubuhi. Karibu sana nyumbani kwetu."

"Mambo vipi? Uko poa?"

Try Soga Today

Soga is now rolling out (in preview), with 6 minutes of access per day for all users around the world. We can't wait to hear your feedback once you try it out.

We are actively working to increase the usage limits and improve the overall experience of Soga.

What's Next

We believe that speech will remain a fundamental interface of the future. It's how humans have communicated for millennia. Soga will be a full release soon, In the coming months we will be making various improvements on both the model and user experience of the app.

We're excited for what's next.

Nadhari AI Lab