BDMUSIC: A Structured Bengali Instrumental Audio Dataset with Attribute-Rich Metadata for Controllable Text-to-Music Generation

Published: 14 July 2026| Version 1 | DOI: 10.17632/z79btczm4f.1
Contributors:
,
,
,
,
,
,
, Raihan Ul Islam

Description

Bengali instrumental music audio dataset for Music Information Retrieval, audio analysis and machine learning purposes - has been developed. BDMUSIC consists of three traditional Bengali instruments, namely, Dotara, Harmonium, and Tabla. The Dotara and Harmonium recordings were collected from live performances, and the Tabla subset was intentionally rendered using a VST-based sampler under predefined musical attributes. Therefore, it can be interpreted as a mixed-acquisition dataset. The dataset has been prepared based on 36 source audio files. Each audio file has been broken down into 30 seconds long audio clips with 32 kHz sampling rate. The dataset consists of 389 audio files with corresponding JSON metadata files and a master CSV metadata file. The metadata includes instrument type, tempo, BPM, genre, mood, musical key, and duration. The dataset can be reused for instrument classification, audio retrieval, music information retrieval (MIR), metadata-driven music analysis, audio representation learning, and controllable text-to-music generation.

Files

Institutions

Categories

Artificial Intelligence in Music, Bangladesh, Music Composition

Licence