Archive
IJMSTA - Vol. 8 - Issue 2 - July 2026
ISSN 2612-2146
Pages:
14
Ragagen: Agentic AI Orchestration for Raga Music Generation
Authors:
Devananda Sreekanth, Sreekanth Gopi, Sreenija Kanugonda, Vaishnavi Pujala, Rakshitha Narasimhaiah
Categories:
Journal
Abstract -
A conversational agent is a dialogue system that understands and generates natural language to communicate with users through multi-turn interactions. Despite advances in conversational agents, few systems explicitly connect evolving conversational emotions with musically coherent and stylistically constrained generation. Furthermore, maintaining emotional continuity, musical coherence, and adherence to traditional melodic structures remains a significant challenge in interactive music generation systems. Indian raga music is a melodic framework in which specific combinations of notes and characteristic phrases are used to evoke distinct moods and emotions. In this paper, we present an agentic AI framework in which specialized agents analyze conversational context and collaboratively formulate context-aware music-generation prompts for downstream music generation. The methodology adopted in this study employs a multi-agent architecture in which specialized agents sequentially analyze conversational context, infer affective states, select appropriate musical constraints, and construct structured instructions for music generation. To develop the prototype, we implemented five specialized agents responsible for affect analysis, raga selection, composition planning, coherence verification, and orchestration; within this architecture, the Affect Analysis Agent is responsible for inferring emotion from conversational text. The experimental prototype included a multi-turn conversational scenario for generating structured music-generation instructions, but the formal evaluation in this paper focused specifically on phrase-level verification and correction. Using paired degraded and compliant Nattai phrases, we evaluated whether the Coherence Verification and Orchestration Agents could detect ornamentation loss and guide repair. Experimental results showed that the proposed framework reliably detected degraded phrases and substantially improved raga-compliance scores after a single correction step. These findings support the use of agentic AI for verification-driven raga phrase repair within a broader affect-aware music-generation architecture, while direct evaluation of emotion inference, conversational adaptation, emotion-to-raga mapping, and audio output remains future work.
Keywords:
Agentic AI, Affective Music Generation, Multi-Agent Orchestration, Indian Raga Music Generation
Download paper
