Meet the 19-year-old who is transforming Indian classical music with AI-driven stem isolation

At the age of 40, Ludwig van Beethoven was deaf, but that did not stop him from composing the iconic Symphony No. 9—a masterpiece that would go on to shape classical music for centuries to come. His brilliance was in his ability, in addition to decades of hard work, to imagine each instrumental layer without ever hearing them physically, proving that music’s essence lies in its building blocks.

Today, we refer to these components as “stems”—the separate parts (vocals, instruments, percussion) that come together to form a complete composition. While every musical genre assembles its stems in unique ways, Indian classical music—known for its seemingly similar-sounding instruments—presents a special challenge when it comes to isolation.

Existing stem splitters are unsuitable for Indian classical music due to their Western focus. 

That’s where Sur comes in.

Built by 19-year-old Shanit Ghose, Sur is an AI-powered stem splitter and sample library designed specifically for Indian classical music (ICM). Unlike generic tools, it aims to capture the nuanced layers of the genre with precision, allowing modern producers to make use of traditional elements into their work easily.

“My vision from day one is to steer the music industry worldwide in the direction of sampling many more niche genres. There’s this whole undiscovered trench of musical genres. I think Indian classical music is one of those, because there’s so many instruments and variety in them,” Ghose tells YourStory. 

Also Read

SaaS for Indian classical music; Bringing tech to textile

Utilising AI

An alumnus of Oberoi International School in Mumbai, Ghose grew up with music as a constant companion, starting with the piano at the age of four and switching to guitar by ten. The idea for Sur took shape during his high school years, when he was the lead guitarist in a band, creating rock covers and experimenting with new tracks. He is currently exploring programs for a bachelor's degree in engineering.

“Whenever I needed a backing track for my guitar solos, I struggled to find a version of the song that excluded the original guitar part. I could sometimes find good tracks with isolated vocals or individual instruments, but most of the time, the available files were incomplete or not usable,” Ghose says. 

Frustrated, he resorted to using audio software to manually remove the guitar sections and record his own solos. This led him to learn more about the concept of stems—the individual layers of a track.

Eventually, Ghose pursued music production, drawing inspiration from his aunt, a Bollywood music composer, who introduced him to Logic Pro X, a music production application.

It was during this time he decided to sample elements from Indian classical music, such as specific melodies and vocal textures. However, Ghose soon realised that no existing tools could isolate stems from Indian classical recordings, making it difficult to recreate those sounds in original tracks.

In simpler terms, stems are individual audio tracks that make up a complete song. Each stem can be identified as a separate element of the music, such as vocals, drums, bass, or melody.

“That limitation affects any music producer or DJ who wants to implement Indian classical music into their work. While there aren’t many people doing it, they face the same issue. There are online audio ‘splitters,’ but those tools are typically trained on databases of pop, rock, or mainstream Western music, so they don’t handle Indian classical very well. So, I thought, ‘Why not build my own database for my needs?’ And that’s how Sur was born,” he explains. 

Launched in early 2024, Sur uses a proprietary AI algorithm trained on over 350 Indian Classical Music tracks, advanced deep learning models for frequency separation, and a curated sample library from independent artists. Ghose primarily used a program called Demucs to process a song and save the instruments as separate files.

“To build the model, I had to create a training library from scratch. I manually separated 350 songs, teaching the model to recognise Indian instruments like the sitar and tabla instead of standard Western ones. I found that while instruments like the guitar and sitar may seem similar, they operate on different frequency ranges. This required extensive fine-tuning to identify peaks and drops, or what to amplify or suppress,” he says. 

In addition to a stem splitter, the platform offers a curated sample library by collaborating with independent artists (not tied to labels), allowing their work to be sampled while giving them access to a diverse library for their own projects.

Previously, Ghose worked on a project at Sarvam AI, an AI startup, as a Voice Activity Detection Engineer for 2-3 months, which further bolstered his understanding of applying AI across various audio contexts.

“While not directly involved with the LLM, my focus was on voice recognition challenges, particularly in handling diverse languages. Using demucs, I adapted the same approach I used for instrument separation—this time training the model on male and female vocal frequency ranges instead of Indian instruments. Over 2-3 months, I fine-tuned the system to identify where average male and female voices lie, improving its ability to detect and process speech accurately,” he says. 

Also Read

From Rihanna to Diljit: The headliners who rocked the music scene in 2024

Future plans

The global music production software market is valued at approximately $7.5 billion, growing at a CAGR of 9.1%. The Indian music software market is projected to reach $500 million by 2029.

Currently bootstrapped, the startup is in talks for a Seed investment, with famous Indian music composer and singer Salim Merchant expressing interest in supporting the initiative. 

“I presented this to Saleem, and we went through every detail together. He offered his honest feedback and agreed to help scale the project—sharing it through his network and building a platform so that everyone in the industry can access it,” claims Ghose. 

"Sur is an incredible innovation by Shanit, blending Indian classical music with modern technology. Its ability to precisely extract classical stems is a game-changer for music producers globally. I genuinely believe in its potential to scale and open up new opportunities for Indian classical music to find its place in mainstream platforms," Salim Merchant tells YourStory.

"Shanit’s vision is inspiring, and I’m happy to share Sur within my network and support its journey. There’s a clear path forward here, and I’m excited to see how Sur shapes the future of music production," he adds.

Sur operates on a subscription model with freemium access to basic stem splitting, standard subscription ($10/month) with access to the curated library, and premium subscription ($20/month) with additional AI features and exclusive samples. 

Sur follows a fair revenue-sharing model, ensuring that contributing artists earn each time their samples are used, along with empowering creators with high-quality, genre-specific tools. 

It aims to partner with platforms like Logic Pro, Ableton, and FL Studio, integrate with global music education programs, and establish itself as the go-to platform for Indian Classical Music (ICM) internationally. 

In terms of competition, local tools like Spleeter lack ICM-specific capabilities, while global platforms like Splice and Tracklib focus on general music sampling but overlook cultural nuances, Ghose says. 

Going ahead, Ghose is planning to explore sampling Russian and Italian music in the coming months. Currently, the platform has 100 users and aims to onboard 10,000+ users within a year.

“This model not only encourages producers to explore new sounds but also helps independent artists gain recognition and steady income. It’s designed to scale as more users join, much like other subscription-based services, but with a clear focus on empowering niche musical genres,” Ghose says.