ViSQA:一个基准数据集和基线模型,用于回答越南语口语问题
Le Trong Minh1,2, Nguyen Duc Thinh2, Nguyen Khanh Tho Loc1,2
1VNU University of Engineering and Technology, Vietnam National University, Hanoi, Vietnam.
PloS one
|January 12, 2026
概括
这项研究介绍了ViSQA,这是越南语口语问答的新基准. 它显示自动语音识别错误显著影响性能,但对口语数据的训练提高了稳定性.
科学领域:
- 自然语言处理自然语言处理.
- 语音识别 语音识别 语言识别
- 机器学习 机器学习
背景情况:
- 语音问答 (SQA) 集成了自动语音识别 (ASR) 和语言理解.
- 现有的SQA基准仅限于像越南语这样的低资源语言.
- 越南SQA缺乏用于研究和开发的标准化数据集.
研究的目的:
- 介绍ViSQA,这是越南语口语问答的第一个基准数据集.
- 允许在不同自动语音识别 (ASR) 转录质量下对SQA模型进行系统评估.
- 促进研究ASR错误对下游语言理解任务的影响.
主要方法:
- 使用文本转语音和ASR管道扩展了UIT-ViQuAD集体.
- 创建了超过13,000个问答对与对齐的口语音频.
- 包括清洁和降噪的音频变体,以进行可靠的评估.
- 进行了5个基于变压器的模型的实验.
主要成果:
- 自动语音识别 (ASR) 错误大大降低了SQA模型的性能 (例如,ViT5 EM: 62.04%至36.30%).
- 在口语转录上的训练模型增强了对ASR错误的稳定性 (例如,ViT5 EM: 36.30%至50.70%).
- 由于转录不准确,表现显著下降.
结论:
- 维斯卡是越南语口语问答系统的严格基准.
- 强调ASR质量对SQA绩效的关键影响.
- 为越南人开发更强大的SQA模型提供了基础.
相关概念视频
Data Collection by Survey
7.6K
The systematic method of obtaining and analyzing accurate information of a population is called data collection. A survey is a standard method of data collection that involves collecting information from a target human population about their experience, opinion, or knowledge of a product, service, or process. The responses are recorded and interpreted. The most common survey examples are written questionnaires, face-to-face or telephonic conversations, focus groups, and electronic (e-mail or...
7.6K
Stratified Sampling Method
11.7K
Sampling is a technique to select a portion (or subset) of the larger population and study that portion (the sample) to gain information about the population. The sampling method ensures that samples are drawn without bias and accurately represent the population. Because measuring the entire population in a study is not practical, researchers use samples to represent the population of interest.
To choose a stratified sample, divide the population into groups called strata and then take a...
To choose a stratified sample, divide the population into groups called strata and then take a...
11.7K
Types of Surveys
468
Surveys are essential for marking property boundaries near water bodies. Different types of surveys are defined, each with its own function. Land surveys mark the property boundaries, while route surveys determine the position of properties on nearby highways. Topographic surveys create maps by capturing the three-dimensional features of the land. Hydrographic surveys focus on the shapes of underwater areas and the movement of streams through the properties. Mine surveys determine the relative...
468
Typical Model Studies
842
Fluid mechanics model studies often utilize scaled-down systems to predict fluid behavior in full-scale environments, such as river flows, dam spillways, and structures interacting with open surfaces. Maintaining Froude number similarity in river models is crucial, as it replicates surface flow features like wave patterns and velocities.
842
Language Development
1.1K
Children master language quickly and with relative ease, supported by both biological predisposition and reinforcement. B. F. Skinner (1957) proposed that language is learned through reinforcement, while Noam Chomsky (1965) argued that language acquisition mechanisms are biologically determined.
The critical period for language acquisition suggests that the ability to acquire language is at its peak early in life. As people age, this proficiency decreases. Language development begins very...
The critical period for language acquisition suggests that the ability to acquire language is at its peak early in life. As people age, this proficiency decreases. Language development begins very...
1.1K
Cochran's Q Test
1.2K
Cochran's Q Test is a nonparametric statistical test used to determine if there are potential differences in the outcomes of three or more related groups on a binary (yes/no) or dichotomous outcome. It is essentially an extension of the McNemar Test, which is limited to two related samples - Cochran's Q test can handle three or more related samples, making it more versatile in scenarios where subjects are measured under multiple conditions. The test statistic follows a Chi-Square...
1.2K

