• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>IEEE Global Conference on Consumer Electronics >An Investigation of End-to-End Speech Recognition Using Model Adaptation for Dysarthric Speakers
【24h】

An Investigation of End-to-End Speech Recognition Using Model Adaptation for Dysarthric Speakers

机译：扰动扬声器模型适应对端到端语音识别的调查

获取原文
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

In this paper, we present an end-to-end automatic speech recognition (ASR) system for dysarthric speech. Because the speaking style of a person suffering from an articulation disorder is quite different from that of a physically unimpaired person, speech recognition systems for such persons need to be constructed in such a way that they specialize in meeting the needs of such dysarthric people. However, the amount of training data that can be collected from dysarthric people is limited because of their large burden. Therefore, it is a challenge to effectively train an ASR model for dysarthric people. In this paper, we introduce a model adaptation approach to train a more accurate model with limited training data, which adapts an ASR model trained by non-dysarthric speech samples for dysarthric speech recognition. From our experiments on an ASR task with two dysarthric subjects, the model adaptation approach with non-dysarthric speech showed better performance than training from scratch.
机译：在本文中，我们介绍了一种用于发育性语音的端到端自动语音识别（ASR）系统。因为患有铰接性障碍的人的讲话方式与身体未受下的人的人完全不同，所以对这些人的语音识别系统需要以这样的方式构建，使他们专注于满足这种缺陷人的需求。但是，由于其负担的巨大负担，可以从缺陷人口收集的培训数据的数量。因此，有效地培训缺陷人的ASR模型是一项挑战。在本文中，我们介绍了一种模型适应方法来培训具有有限训练数据的更准确的模型，这适应了由非发育性语音样本训练的ASR模型进行扰乱语音识别。从我们的ASR任务的实验从两个发狂主题的ASR任务中，具有非发育性语音的模型适应方法显示出比从头划痕的训练更好的性能。

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