VIVOS: Vietnamese Speech Corpus for ASR

VIVOS Corpus VIVOS is a free Vietnamese speech corpus consisting of 15 hours of recording speech prepared for Automatic Speech Recognition task. The corpus was published by AILAB, a computer science lab of VNUHCM - University of Science, with Prof. Vu Hai Quan is the head of. We publish this corpus in hope to attract more scientists to solve Vietnamese speech recognition problems. The corpus should only be used for academic purposes. License Creative Commons Attribution NonCommercial ShareAlike v4.0 (CC BY-NC-SA 4.0) (details) Associated paper Please cite this paper when using VIVOS corpus for research "A non-expert Kaldi recipe for Vietnamese Speech Recognition System", Hieu-Thi Luong and Hai-Quan Vu, in Proc. WLSI/OIAF4HLT2016 (paper) Contact [e-mail] Corpus properties Speech was recorded in a quiet environment with high quality microphone, speakers were asked to read one sentence at a time. Training Testing Speakers 46 19 Male 22 12 Female 24 7 Utterances 11660 760 Duration 14:55 00:45 Unique Syllables 4617 1692 Evaluations The corpus was evaluated using our non-expert recipe for Vietnamese Speech Recognition system which is described in the associated paper. baseline +pitch +tone mGMM 19.66 15.14 14.91 mGMM+MMI 18.08 14.96 13.91 mGMM+SAT 15.79 12.07 12.13 mDNN+SAT 13.34 9.54 9.48 Notice This is the official replacement for http://ailab.hcmus.edu.vn/vivos/

VIVOS语料库是一款免费的越南语语音语料库，总时长15小时，专为自动语音识别（Automatic Speech Recognition）任务打造。该语料库由胡志明市自然科学大学（VNUHCM - University of Science）计算机科学实验室AILAB发布，实验室由武海泉（Vu Hai Quan）教授担任负责人。我们发布本语料库旨在吸引更多科研人员投身越南语语音识别相关研究，且本语料库仅可用于学术研究用途。
许可协议：知识共享署名-非商业性使用-相同方式共享4.0国际许可协议（Creative Commons Attribution NonCommercial ShareAlike v4.0，CC BY-NC-SA 4.0）（详情）
关联论文：若将VIVOS语料库用于科研，请引用以下论文：《A non-expert Kaldi recipe for Vietnamese Speech Recognition System》，作者为Hieu-Thi Luong与Hai-Quan Vu，发表于Proc. WLSI/OIAF4HLT2016（论文链接）
联系方式：[e-mail]
语料库属性：语音录制于安静环境，采用高品质麦克风采集；要求朗读者逐句朗读文本。
语料库参数：
- 总说话人数：训练集46人，测试集19人；其中男性训练集22人、测试集12人，女性训练集24人、测试集7人
- 语句总数：训练集11660条，测试集760条
- 总时长：训练集14小时55分，测试集45秒
- 唯一音节数：训练集4617个，测试集1692个
评估情况：本语料库已通过关联论文中提及的越南语语音识别非专家训练配方完成评估，测试结果如下：
| 模型配置 | 基准性能 | 加入音高特征 | 加入声调特征 |
| ---- | ---- | ---- | ---- |
| mGMM | 19.66 | 15.14 | 14.91 |
| mGMM+MMI | 18.08 | 14.96 | 13.91 |
| mGMM+SAT | 15.79 | 12.07 | 12.13 |
| mDNN+SAT | 13.34 | 9.54 | 9.48 |
注意：本语料库为原官网http://ailab.hcmus.edu.vn/vivos/ 的官方替代版本。

搜集汇总

背景与挑战

背景概述

VIVOS是一个免费的越南语语音语料库，专为自动语音识别任务设计，包含15小时的录音语音，由AILAB实验室发布。该数据集包括训练集和测试集，涵盖65位说话者、12420个话语和6309个独特音节，适用于学术研究，并遵循CC BY-NC-SA 4.0许可证。