The Performance Comparison between Speech Recognition Tool and Text Interpreting when Class Archives are made with Subtitle for Communication Access

我们可以从与我们合作研究的聋哑学生那里得到报告。它是通过重复输入到语音识别引擎的系统的关于错误识别和错误转换的数据。此外，在群马大学金泽实验室的支持下，对文本翻译的替代打字方法和重复型语音识别方法进行了并行实验。至于文本翻译和语音识别，我们通过在美国的调查了解到，它们作为业务逐渐分离。目前，在日本，语音到文本服务还没有实现为商业。然而，这是在高等教育中实现连续语音到文本服务时必须考虑的问题。我们在科学研究补助金的支持下进行了研究，取得了以下成果。(1)在拍摄系统中设置语音识别引擎（Dragon Naturally Speaking）作为视频同步字幕成为可能。(2)在比较语音文本转换系统的实用程度时，目前最常用的是两人或多人交替输入的文本翻译系统（IPTalk）。(3)当我们比较两种训练的字幕员的文本翻译和重复/纠正的人，前者是有利的。这是因为文本口译设备简单，也不需要专门的房间。然而，语音识别系统的操作员的劳动力成本高于美国的文本翻译。这是因为许多人认为语音识别系统的操作是一项“困难的工作”。"

We could get the reports from the deaf student who cooperated in our research. It is the data about incorrect recognition and incorrect conversion by the system which inputs into the voice recognition engine by repetition. Moreover, supported by Kanazawa laboratory in Gumma University, the alternative typing method for Text interpreting was experimented in parallel with the method of repetition type speech recognition.As for Text interpreting and speech recognition, we appreciated that they were separated gradually as business, by investigation in the U.S.A. Each customer group is being divided. Now, in Japan, Speech-to-Text service is not materialized as business. However, this is the subject which must surely be considered, when realizing continuous Speech-to-Text service in higher education.We were able to get the following results in our research supported with the grants-in-aid for scientific research.(1) It became possible to set a speech recognition engine (Dragon Naturally Speaking) into the capturing system as a video-synchronous subtitle.(2) Text interpreting system (IPTalk) operated by two or more person's alternative typing is most useful at present when we compare degree of practical use in some Speech-to-Text Systems.(3) When we compare two trainings for captionists of Text interpreting and for repetition/correction person, the former one is advantageous. It is because Text interpreting is simple for equipment and the special room is not needed, either. However, the operator of a speech recognition system has a labor cost higher than Text interpreting in the U.S. which precedes. This is because many people think that operation of speech recognition system is "difficult work."