Digital Umuganda’s Kinyarwanda NLP in Rwanda

Most complicated and intricate endeavors are not necessarily the best and most adopted. Having a bank account requires going through a process that’s exponentially slower than getting financial services on mobile phones. People are more likely to have a phone number than a list of documents and time to show up in a physical bank.

Adding an additional layer of AI-based solution like chatbot, voice assistant and facial recognition for KYC process is what really takes it to whole new level. Furthermore these technologies are timeless and borderless making it a global solution. Companies who understand these key lessons are unstoppable.

Google was ahead of this game for about a decade and in 2017 its data science events in Tanzania and South Africa led it to believe that being at the avant garde in AI in Africa is going to be pay huge dividends. It opened its AI research facility in Ghana in 2018 to find AI based solutions for population planning, humanitarian response, environmental protection, statistical indicators and vaccination planning among other things.

The only way to get all this data however is to learn myriad of languages spoken in Africa. To use NLP with the different languages and build intelligent models they needed to get data from multifarious data sources and that isn’t an easy task.

Luckily its not only google that comes up with great ideas. Digital Umugandas success in collecting Kinyarwanda Voice and text datasets is evident through Mozilla Common Voice’s platform With over 2000 hours collected & curated. The volume of data in Kinyarwanda is second only to English, the largest open voice dataset in the world. This was achieved by building productive partnerships with local communities on the ground.

Company is lead by Audace Niyonkuru who studied in China and studied the Alipay, WeChat and SuperApp systems and came up with his own ideas that he kept working on along with some other very smart colleagues that he is working with. It is an immense pleasure to work with Audace and his team to brainstorm on applied use cases for its application and mass adoption.

A free version can be found on Digital Umuganda’s Github page for open access. It has common voice datasets, chatbot setup instructions and Speech to Text engine. DeepSpeech is an open source Speech-To-Text engine, using a model trained by machine learning techniques based on Baidu’s Deep Speech research paper. DeepSpeech uses Google’s TensorFlow to make the implementation easier.

With other major languages spoken in african continent it will open the continent to major players in a diverse industries to expand their operations using local langauges on their digital platforms with minimal investment. That opens up horizon for companies with access to second most largest and populous continent.

So excited to be part of something so powerful and impactful that has the potential to change many lives.

https://github.com/Digital-Umuganda/Deepspeech-Kinyarwanda