Gurudeva's Voice, Reborn

It's something we didn't foresee happening this soon, or perhaps not at all.

Our founder, the late Satguru Sivaya Subramuniyaswami, always intended to record himself speaking the edited and published version of his Master Course book trilogy lessons comprising three 1,008-page books--Dancing with Siva, Living with Siva and Merging with Siva. However, his life of service and sadhana was quite busy, and taking the time to properly audio record already-published material was not a top priority. In the final years of his life, he recorded over a hundred lessons from Merging with Siva. He had the monks set up a high-end recording station and every afternoon he would spend an hour or two reading the lessons. He did this for some months. And then, rather suddenly in 2001, after returning from taking 72 Innersearchers through Northern Europe, he learned that it was time to release the physical body. Siva's Will Be Done.

We created an audio CD of a few chapters from those Merging with Siva recordings, and contented ourselves with that and Gurudeva's other spontaneous audio talks for a number of years. Now and then we attempted to replicate his voice, but no one could make it work. So we waited.

Just last year, AI capabilities took off. Over a year ago we saw articles popping up about voice cloning AI systems. Feeling that it would be good to research a potential fulfillment of Gurudeva's original intention, we looked into it. After trying out a few AIs, we found the ElevenLabs model to be quite promising. It wasn't excellent at the time, but their model has been evolving rapidly, and the results now are quite impressive. We have been able to replicate Gurudeva's voice to the point that it sounds so accurate those who hear it and who knew him are stunned.

This is only possible because, fortunately, we have good quality recordings of Gurudeva speaking over 100 Merging with Siva lessons. We fed a couple hours of those recordings into the AI. We now have a voice clone that can speak any block of text that we feed into it. Though the voice is technically artificial, we feel the shakti is still present because it is created with the exact tones and style of his original voice.

We plan to gradually create audio files of all the Master Course trilogy lessons, linking them in the web pages and daily lesson emails. Then we'll continue with Gurudeva's other books.

One challenge is the use of Sanskrit words sprinkled amongst English. The AI English-language model we use is not able to recognize diacritical marks or long and short vowels to pronounce all the Sanskrit words correctly. Fortunately, ElevenLabs allows us to upload a lexicon text file of individual word replacements. When the AI is voicing a block of text, it checks this file to see if the spelling of any particular word in the block is changed, and then speaks out the changed version. We change the spelling to make it sound more phonetically correct. For example, when the word "jñānī" appears in a Merging with Siva lesson, we change the spelling to "yaanee," and then the AI pronounces it correctly. This is a slow process, as we have to experiment with various spellings until the AI gets it right. We are gradually building the lexicon now.

We are firmly committed to ensuring that this digital voice is used exclusively for reproducing what Gurudeva actually wrote or said. Crossing this line would not only be confusing but also unjust to all parties involved, including Gurudeva himself.

Enough theory. Below, we are pleased to share a sample of Gurudeva's digital voice reading from Merging with Siva, lesson one. As the AI model gets more refined, we expect the quality to increase even more.

[audio mp3="https://www.himalayanacademy.com/taka-archive-all-assets/wp-content/uploads/2024/04/MwS-Lesson-1.mp3"][/audio]