My Journey into Artificial Intelligence through Seva

Language barriers can often be challenging. Most of the self-development content during my seva with SRMD were in Gujarati. This made it challenging for the mission to reach out to a wider international audience. So, I embarked on a mission to make  SRMD’s online content  more accessible. 

May 2021: Manual Work

For a couple of years, the Gujarati content was being translated manually into English, Spanish, and French using an online program, Maestra. Over time, production expanded to translate five more languages: German, Hindi, Mandarin, Portuguese, and Russian, requiring more manpower.

To address the issue of people struggling to read subtitles quickly for long periods of time, artists joined the team to dub voice-overs  in eight languages. This increased production times and effort. As the seven YouTube channels went live, viewership skyrocketed, resulting in our output growing multifold. The group was tasked with producing over fifty videos a month, instead of just ten!

July 2023: AI Integration

The process remained resource-intensive, so we began considering the use of technology to reduce human efforts. For example, we created an automated backend system (using Java) that manages files and archives by changing file directories to different folders as they pass the stages of production. A translation file would be transferred to the dubbing stage, followed by reaching the video editor’s folder, and finally to QC. Sevaks were no longer required to spend hours organizing  hundreds of video files across a massive database. This improved efficiency, but we had more potential. As enthusiasts of AI, we started to research the implementation of a voice synthesis AI model to replicate the speaker’s voice in each of the eight languages. This would save a lot of time and would spare the tedious manual work.  

I was very excited to join the team of sevaks. Working on the AI voice model project, I learned from programming experts. We built a language-specific model based on Open-AI’s text-to-speech API. The model was trained using a text-to-voice dataset, with inputs of an ‘SRT’ subtitles file, and outputs of SRMD speaker’s content. A functional model was achieved after employing multiple data post-processing techniques. This also took us hundreds of hours of training. These months of seva were not just volunteering work for me, but a journey of personal skill development along with walking into the realm of AI.  

January 2024: Using AI to Achieve Success

Our group set out to maximize the impact of Artificial Intelligence by successfully implementing the model into video production. The first step was to eliminate unnecessary human effort. We set up specialized  platforms like Teams and Trello and have developed a systematic production chain across coordinators. This ensured accuracy in translation, quality checks, and timely uploads.

Integrating the AI program significantly increased our production capacity and viewership.  Expert counsel states that we are ‘two notches’ ahead of market technology while incurring only a fifth of the cost, saving hundreds of thousands of dollars. This is crucial for an NPO like SRMD with limited financial resources.

Now, we look to monetize our channels through advertising and subscriptions. Merchandise sales would also provide funding for  sustaining our AI program.

Today: Operation Successful

The impact of this endeavor  is profound. The seven language channels have reached over 30 million views across YouTube, Facebook, and Instagram since their inauguration in 2021. Over 15% of SRMD’s 2.5 million social media following is from foreign-language channels, facilitating the increased access of a global audience to self-development content.

By being an indispensable part of this project, I have learned valuable technical and business skills.