Nepali Text To Speech project started back in 2015 by Kaushal Subedi (me). When the tragic earthquake shook the whole country and caused a lot of damage, I was here in the United States and could not do much to help with the crisis relief efforts directly. I got in contact with a non-profit organization called Translators Without Borders and found out that a lot of international organizations had sent people out there in Nepal to help with the relief, but one of the biggest issues they were having was communicating with the locals in remote parts of Nepal where most people did not understand anything other than Nepali. I started helping Translators Without Borders with translation of phrases that could be used by the rescue teams, since I had previous translation experience here in the United States as well.
After a few days it hit me that we could leverage technology to help with this situation. I had background with developing text to speech applications for Android platform when the platform was new and the feature was not a part of the core system. I found out that Bhasha Sanchar which is now Language Technology Kendra, a non profit organization from Nepal had started efforts to create a text to speech system in Nepali which was the only one of it’s kind around 2006, but the project never got completed. They did however, open sourced what they had on their website. So I took the source code and started a 16 hour binge to create the first iteration of this application. It used Google Translate to allow rescuers to type in any language that was supported by Google Translate, and it would output a spoken version of that in Nepali. This helped the rescue efforts immensely and made me feel proud that I was able to help in the tragic situation even though I could not physically be there.
Even though I was not expecting anything back from this project, I got thanked by a lot of people and even got invited to talk in some software development conferences in Nepal. Indirectly, I have gotten more than I could have asked for back from this project. I kept maintaining the application and even though rescuers stopped using the application after the rescue efforts were over, there were a lot of other people that still wanted to use the application, either for fun or other purposes. The traffic never died down so I continued to refine the application and changed the UI a little. I took the old version down recently as I was working on a new version that was not based on previous work done by Bhasha Sanchar. The current version uses a different TTS engine which is based on the work done by Carnegie Melon University. I am exploring new ways to make it even better by using machine learning based text to speech approaches, and have been experimenting with Google’s WaveNet based engine.