Vietnamese Multi-Dialect (ViMD) dataset
收藏资源简介:
越南多方言数据集(ViMD)是由越南胡志明市信息技术大学信息科学与工程学院创建的,旨在捕捉越南63个省的方言多样性。该数据集包含102.56小时的音频,约19,000条语音记录,转录文本超过120万字。数据集的创建过程包括视频收集、音频提取、半自动标注和严格的手动验证。ViMD数据集主要用于方言识别和语音识别任务,旨在解决越南多方言语音数据的识别挑战,特别是地理因素对方言的影响和当前语音识别方法在处理多方言数据时的局限性。
Vietnamese Multi-dialect Dataset (ViMD) was created by the School of Information Science and Engineering, Ho Chi Minh City University of Information Technology, Vietnam, with the aim of capturing the dialect diversity across Vietnam’s 63 provinces. This dataset contains 102.56 hours of audio, approximately 19,000 speech recordings, and over 1.2 million words of transcribed text. The dataset construction process includes video collection, audio extraction, semi-automatic annotation, and rigorous manual verification. Primarily intended for dialect identification and speech recognition tasks, ViMD aims to address the challenges in recognizing Vietnamese multi-dialect speech data, particularly the impact of geographic factors on dialects and the limitations of current speech recognition methods when processing multi-dialect datasets.
ViMD数据集概述
数据集简介
ViMD数据集(Vietnamese Multi-Dialect)是一个全面的数据资源,旨在捕捉越南63个省级方言的多样性。该数据集与论文《Multi-Dialect Vietnamese: Task, Dataset, Baseline Models, and Challenges》相关联。
数据来源
数据集的来源是越南63个省的广播电台的新闻节目。
数据集统计
- 总体统计:
- 统计图表:
- 统计图表:
数据集属性
| 键 | 描述 |
|---|---|
| set | 音频集:{train, valid, test}。 |
| filename | 文件名遵循语法 {province code}_{Sequence Number of Audio}。 |
| text | 音频的转录文本。 |
| length | 音频的长度(以秒为单位)。 |
| province | 省级方言代码。 |
| region | 区域方言:{North, Central, South}。 |
| speakerID | 说话人识别代码遵循语法 spk_{province code}_{Sequence Number of Speaker}。 |
| gender | 说话人性别(0代表女性,1代表男性)。 |
引用
如果使用该数据集进行研究,请引用以下内容: bibtex @misc{vandinh2024multidialectvietnamesetaskdataset, title={Multi-Dialect Vietnamese: Task, Dataset, Baseline Models and Challenges}, author={Nguyen Van Dinh and Thanh Chi Dang and Luan Thanh Nguyen and Kiet Van Nguyen}, year={2024}, eprint={2410.03458}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2410.03458}, }




