• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文期刊>電子情報通信学会技術研究報告. 音声. Speech >Evaluation of Speaker Model Selection based on Bayesian Information Criterion in Unsupervised Speaker Indexing
【24h】

Evaluation of Speaker Model Selection based on Bayesian Information Criterion in Unsupervised Speaker Indexing

机译：基于贝叶斯信息准则的无监督说话人索引中说话人模型选择的评估

获取原文
获取原文并翻译 | 示例
           
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

This paper addresses unsupervised speaker indexing for discussion audio archives. We have performed the speaker indexing using our proposed framework that selects an optimal speaker model (GMM or VQ) based on the BIG. A threshold of the speaker indexing is needed to be determined in advance because the framework is applied to the speaker indexing in the case where the number of speakers is unknown beforehand. Thus, we evaluate robustness of indexing accuracy when varying the threshold and the indexing accuracy when the number of speakers instead of the threshold is given. As a result of comparison with conventional methods, it is demonstrated that the proposed framework can set up the threshold robustly and archives the higher indexing accuracy in both cases where the number of speakers is unknown or given beforehand. The speaker index is useful for speaker adaptation of the acoustic model, which improves the performance of automatic speech recognition.
机译：本文讨论了讨论音频档案的无监督发言人索引。我们使用建议的框架执行了发言人索引，该框架根据BIG选择最佳的发言人模型（GMM或VQ）。因为在事先未知说话者的数量的情况下将框架应用于说话者索引，所以需要预先确定说话者索引的阈值。因此，我们在改变阈值时评估索引准确性的鲁棒性，而在给出扬声器数量而不是阈值时评估索引准确性。与常规方法进行比较的结果表明，在说话者数量未知或事先给定的两种情况下，所提出的框架都可以稳固地设置阈值并存档更高的索引准确性。说话者索引可用于声学模型的说话者自适应，从而提高自动语音识别的性能。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号