Adaptive AI-driven Music Recommendation for Small-Scale Studios

Thesis Abstract

This study investigates how adaptive AI-driven music recommendation systems can be optimized for small-scale recording studios to enhance creative workflow, reduce time-to-first-mix, and improve track coherence across diverse projects. The problem addressed is the misalignment between generic streaming-oriented recommender models and the nuanced, context-specific needs of small studios handling limited budgets, diverse genres, and evolving artist rosters. The aim is to develop an adaptive recommendation framework that leverages domain-specific metadata, project phase signals, and user feedback to deliver contextualized track suggestions that align with production goals, sonic branding, and collaborative decision-making. Specific objectives are (1) to identify key variables influencing music recommendations in small-studio production pipelines; (2) to design a hybrid recommender architecture integrating content-based, collaborative, and domain-adaptive components; (3) to implement a lightweight, privacy-preserving prototype validated in real-world studio sessions; (4) to evaluate system performance against baseline recommendations using multi-criteria metrics including relevance, diversity, and user satisfaction; and (5) to formulate guidelines for deployment in resource-constrained environments. The methodology adopts a mixed-methods research design combining quantitative evaluation of recommender performance with qualitative insights from music producers. The population comprises 60 professional producers and engineers from 20 small-scale studios across three metropolitan regions. A purposive sample of 40 participants is selected for the quantitative evaluation, with 12 studios providing longitudinal usage data over a 12-week pilot. Data collection instruments include (a) a structured usability and satisfaction survey (Likert-scale items, n=320 responses), (b) studio session logs capturing project type, genre, tempo, key, and production stage, and (c) a bespoke recommendation logger recording suggested tracks, metadata, user actions, and time-to-decide. Instruments are validated through content validity checks with two domain experts and a pilot test (n=10). The adaptive recommender model employs a hybrid architecture that fuses (i) content-based features (bpm, key, genre, instrumentation), (ii) collaborative signals (co-producer interactions, playlist co-editing), and (iii) domain-adaptive weighting informed by a reinforcement-learning-like meta-parameter adjusting through Bayesian optimization. The theoretical lens integrates Legitimation Theory to interpret stakeholder acceptance in practice and the Technology Acceptance Model (TAM) extended with perceived contextual usefulness. Analytical techniques include regression analysis to quantify attribute effects on relevance scores, MAUT-based multi-criteria decision analysis to optimize trade-offs between relevance and diversity, and mixed-effects modeling to account for nested studio-level variations. The qualitative component uses thematic analysis of producer interviews (n=24 interviews) to extract themes around workflow integration, creative autonomy, and trust in AI suggestions. Expected findings anticipate that adaptive recommendations incorporating project-phase signals and domain-adaptive weights will significantly improve relevance (mean increase of at least 15%), diversity (15–20% broader stylistic coverage), and time-to-decision (reduction of 18–25%) compared with baseline non-adaptive recommendations. The study also expects nuanced variances by genre and production stage, with more pronounced benefits in multi-genre projects and late-stage mixing sessions. The findings will elucidate how producers interpret AI-suggested tracks within established creative processes, revealing factors that facilitate or hinder adoption, such as perceived controllability, explainability of suggestions, and alignment with sonic branding. Contribution to knowledge includes (1) a novel hybrid adaptive recommender tailored for small-studio production workflows, (2) a validated methodological framework for evaluating AI-assisted creative tools in resource-constrained settings, (3) empirical evidence on how domain-adaptive weighting and session-level signals impact usability and decision quality, and (4) practical deployment guidelines addressing data privacy, minimal-footprint computation, and stakeholder governance. The main conclusion posits that context-aware adaptive AI recommendations can meaningfully augment creative decision-making in small-scale studios when designed with explicit workflow integration, user control, and transparent feedback loops. The study recommends (a) iterative pilot deployments coupled with ongoing user training, (b) incorporation of explainable AI components to enhance trust, (c) scalable architectures that accommodate studio growth and genre expansion, and (d) continued evaluation of long-term impact on creative outcomes and project turnaround times.