PerFormer: An AI-Driven Approach to Melody Generation in Microtonal Persian Music

Farzad Hosseinabadi, Ian Gibson, and Christopher Dewey

Proceedings of the International Conference on New Interfaces for Musical Expression

Abstract

This paper presents PerFormer, a Transformer-based framework for generating monophonic microtonal melodies in Persian classical music. While most symbolic music generation systems are designed for the twelve-tone equal temperament (12-TET) system, many musical traditions employ microtonal pitch structures that fall outside this framework. Persian classical music is a prominent example, characterized by monophonic textures and distinctive intervallic systems that are not adequately supported by existing generative models.To address this gap, we introduce a culturally informed event-based symbolic representation that encodes microtonal pitch categories, including Persian accidentals such as sori (quarter-tone sharp) and koron (quarter-tone flat), together with rhythmic duration and intra-measure positional information. The model is trained on a curated dataset of Persian melodies, augmented through controlled microtonal transposition, and represented as position-pitch-duration token sequences.The system is evaluated using both quantitative metrics and a perceptual listening study in comparison with a first-order Markov baseline. Quantitative results show that the model preserves modal pitch inventories, rhythmic duration characteristics, and metric position distributions, while perceptual evaluation indicates substantial improvements in musical coherence, stylistic similarity, pitch correctness, and overall quality.These findings indicate that Transformer architectures can be effectively adapted to microtonal musical traditions when supported by appropriate representational design. The proposed framework provides a foundation for future work in computational tools for improvisation, composition, and pedagogy within culturally diverse musical systems.

Citation

Farzad Hosseinabadi, Ian Gibson, and Christopher Dewey. 2026. PerFormer: An AI-Driven Approach to Melody Generation in Microtonal Persian Music. Proceedings of the International Conference on New Interfaces for Musical Expression. DOI: 10.5281/zenodo.20784117 [PDF]

BibTeX Entry

@inproceedings{nime2026_30,
 abstract = {This paper presents PerFormer, a Transformer-based framework for generating monophonic microtonal melodies in Persian classical music. While most symbolic music generation systems are designed for the twelve-tone equal temperament (12-TET) system, many musical traditions employ microtonal pitch structures that fall outside this framework. Persian classical music is a prominent example, characterized by monophonic textures and distinctive intervallic systems that are not adequately supported by existing generative models.To address this gap, we introduce a culturally informed event-based symbolic representation that encodes microtonal pitch categories, including Persian accidentals such as sori (quarter-tone sharp) and koron (quarter-tone flat), together with rhythmic duration and intra-measure positional information. The model is trained on a curated dataset of Persian melodies, augmented through controlled microtonal transposition, and represented as position-pitch-duration token sequences.The system is evaluated using both quantitative metrics and a perceptual listening study in comparison with a first-order Markov baseline. Quantitative results show that the model preserves modal pitch inventories, rhythmic duration characteristics, and metric position distributions, while perceptual evaluation indicates substantial improvements in musical coherence, stylistic similarity, pitch correctness, and overall quality.These findings indicate that Transformer architectures can be effectively adapted to microtonal musical traditions when supported by appropriate representational design. The proposed framework provides a foundation for future work in computational tools for improvisation, composition, and pedagogy within culturally diverse musical systems.},
 address = {London, United Kingdom},
 articleno = {30},
 author = {Farzad Hosseinabadi and Ian Gibson and Christopher Dewey},
 booktitle = {Proceedings of the International Conference on New Interfaces for Musical Expression},
 doi = {10.5281/zenodo.20784117},
 editor = {Benedict Gaster and João Tragtenberg and Anna Xambó and Tom Mitchell},
 issn = {2220-4806},
 month = {June},
 note = {},
 numpages = {6},
 pages = {257--262},
 presentation-video = {https://youtu.be/uHZb1RAoEZ8},
 title = {PerFormer: An AI-Driven Approach to Melody Generation in Microtonal Persian Music},
 track = {Paper},
 url = {http://nime.org/proceedings/2026/nime2026_30.pdf},
 urlsuppl1 = {http://nime.org/proceedings/2026/nime2026_30_file01.wav},
 year = {2026}
}