Multi-Accent Mandarin Dry-Vocal Singing Dataset: Benchmark for Singing Accent Recognition

Multi-Accent Mandarin Dry-Vocal Singing Dataset: Benchmark for Singing Accent Recognition Zihao Wang Zhejiang University Hangzhou, China Carnegie Mellon University Pittsburgh, United States carlwang@zju.edu.cn Shulei Ji Zhejiang University Hangzhou, China Innovation Center of Yangtze River Delta, Zhejiang University Hangzhou, China shuleiji@zju.edu.cn Le Ma Zhejiang University Hangzhou, China maller@zju.edu.cn Yuhang Jin Zhejiang University Hangzhou, China 3210103422@zju.edu.cn Shun Lei Tsinghua University Shenzhen, China leis21@mails.tsinghua.edu.cn Jianyi Chen Hong Kong University of Science and Technology Hongkong, China jchenil@connect.ust.hk Haoying Fu Mei KTV Beijing, China 326452438@qq.com Roger B. Dannenberg Carnegie Mellon University Pittsburgh, United States rbd@cs.cmu.edu Kejun Zhang∗ Zhejiang University Hangzhou, China Innovation Center of Yangtze River Delta, Zhejiang University Hangzhou, China zhangkejun@zju.edu.cn Abstract Singing accent research is underexplored compared to speech ac- cent studies, primarily due to the scarcity of suitable datasets. Ex- isting singing datasets often suffer from detail loss, frequently resulting from the vocal-instrumental separation process. Addi- tionally, they often lack regional accent annotations. To address this, we introduce the Multi-Accent Mandarin Dry-Vocal Singing Dataset (MADVSD). MADVSD comprises over 670 hours of dry vocal recordings from 4,206 native Mandarin speakers across nine distinct Chinese regions. In addition to each participant record- ing audio of three popular songs in their native accent, they also recorded phonetic exercises covering all Mandarin vowels and a full octave range. We validated MADVSD through benchmark ex- periments in singing accent recognition, demonstrating its utility for evaluating state-of-the-art speech models in singing contexts. Furthermore, we explored dialectal influences on singing accent and analyzed the role of vowels in accentual variations, leveraging MADVSD’s unique phonetic exercises. ∗Corresponding author Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than the author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions from permissions@acm.org. MM ’25, Dublin, Ireland © 2025 Copyright held by the owner/author(s). Publication rights licensed to ACM. ACM ISBN 979-8-4007-2035-2/2025/10 https://doi.org/10.1145/3746027.3758210 CCS Concepts • Applied computing →Sound and music computing; • Com- puting methodologies →Speech recognition; • Information systems →Multimedia databases. Keywords Mandarin Singing Dataset; Singing Accent Recognition; Dry Vocals; Regional Chinese Accents ACM Reference Format: Zihao Wang, Shulei Ji, Le Ma, Yuhang Jin, Shun Lei, Jianyi Chen, Haoying Fu, Roger B. Dannenberg, and Kejun Zhang. 2025. Multi-Accent Mandarin Dry-Vocal Singing Dataset: Benchmark for Singing Accent Recognition. In Proceedings of the 33rd ACM International Conference on Multimedia (MM ’25), October 27–31, 2025, Dublin, Ireland. ACM, New York, NY, USA, 8 pages. https://doi.org/10.1145/3746027.3758210 1 Introduction In recent years, speech and music models have developed rapidly [8, 10, 25, 34, 36, 38, 42]. However, in the field of accent research, the singing domain significantly lags behind the speech domain. This is primarily due to the lack of high-quality accent datasets in the singing field [17]. Existing large-scale singing datasets are mostly wet vocal tracks extracted from songs [4]. Audio effects processing (such as reverb, echo, equalization, etc.) inevitably sacrifices original sound details. Furthermore, existing dry vocal singing datasets [16, 45] generally lack regional accent labels. To bridge this data gap and facilitate focused research on singing accent, we introduce the Multi-Accent Mandarin Dry Singing Vocal arXiv:2512.07005v1 [cs.SD] 7 Dec 2025 MM ’25, October 27–31, 2025, Dublin, Ireland Zihao Wang et al. Dataset (MADVSD)1. MADVSD is a relatively large-scale, metic- ulously curated dataset comprising over 670 hours of dry vocal recordings from 4,206 native Mandarin speakers across diverse geographical regions of China. Collaborating with a nationwide organization, their employees across various provinces and cities were coordinated to record dry vocals at home using their mobile phones. These participants were recorded performing popular Man- darin songs and specifically designed phonetic exercises. These exercises encompass all Mandarin vowels and a full octave range of scales, providing granular phonetic and acoustic data for accent analysis. The data

This content is AI-processed based on open access ArXiv data.