MUMBAI, India, May 1 -- Intellectual Property India has published a patent application (202611027420 A) filed by Manipal University, Jaipur, Rajasthan, on March 9, for 'system and method for personalized music resource recommendation based on audio embeddings and contextual metadata.'

Inventor(s) include Dr. Shivendra Dubey; Sourya KVS; Arnav Malik; Aryaman Jaitley; Yashvit Kumar; Arshita Sinha; Vaishnavi Srivastava; and Sakshi Dubey.

The application for the patent was published on May 1, under issue no. 18/2026.

According to the abstract released by the Intellectual Property India: "The present invention relates to a recommender system that is designed specially for musicians of every skill type, it focuses on providing personalised resources for enhanced learning and better practice. The system comprises a data acquisition module to collect and process music from datasets such as such as Jamendo, FMA, and MagnaTagATune; feature extraction and embedding generation module analyze the audio input of the user by using Librosa and Essentia libraries and generates corresponding textual and semantic embeddings by using Gemini embeddings or SentenceTransformers; vector database and similarity mapping module resulting embeddings are stored and indexed within a vector database through a vector mapping module to enable efficient similarity searches; Clustering and Genre Representation module applies unsupervised machine learning algorithms, including K-Means and DBSCAN, to organize embeddings into meaningful clusters representing musical styles, genres, and skill-related patterns; and lightweight FastAPI backend oversee the entire pipeline. It will take audio inputs, call embedding models, look for similarities in the vector storage, and provide ranked recommendations. The present platform analyses users own playing style by actually understanding their play style by taking their videos as input further matching it with contextual metadata taken from open source music datasets and YouTube. This invention uses pretrained models which convert inputs into audio embeddings, vector similarity search and clustering algorithms to smoothly identify similar music generating targeted recommendations."

Disclaimer: Curated by HT Syndication.