Developping a system for dialog support to the hard of hearing via speech recognition and character display

开发一个通过语音识别和字符显示为听力障碍人士提供对话支持的系统

• 批准号: 18500127
• 负责人: FUNADA Tetsuo
• 金额: $ 0.82万
• 依托单位: Kanazawa University
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for Scientific Research (C)
• 财政年份: 2006
• 资助国家: 日本
• 起止时间: 2006 至 2007
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-18500127/
• 关键词: the hard of hearing; dialog support system; speech recognition; noisy environment; reliability

This research is to develop a dialog support system for persons such as age-related hearing loss via speech recognition and character display. For this purpose, the system is expected to be small and light. At the beginning of this research, we intended to design the system with combining custom ISI and micro-displays. However since mobile computers with a built-in microphone recently have come onto the market, we changed to implement the system on a mobile computer.At the beginning in 2006, the system was designed to recognize only the pre-registered words, so it shows inconvenience that unregistered words cannot be recognized. Therefore, we improved the system by making it recognize all Japanese 101 syllables (kana). In this case, dialog partner has to speak syllable by syllable and the system display the recognized syllables one after another. It has also inconvenience that the system can display contents of dining by a sequence of only kana character so difficult to read fir the haul of hearing.In 2007, we improved the system in the following three points. The first is to decrease recognition error by incorporating new algorithm to extract noise-robust speech feature proposed by us. The second is to incorporate the "kana to Chinese character conversion" in the system, the command of which is also ordered by speech The third is to make the system display the ascend candidate of recognition, and can select the second one in case of mis-recognition of the first candidate. This selection command is also acceptable by speech. By making the system be possible to operate only by speech command, it has become convenient to use.Further improvement is desired in recognition speed and recognition error.Parts of these results have been presented at technical committee on speech of Acoustic Society of Japan and the Institute of Electronics, Intimation and Communication Engineers, and International Workshop on Communication and Information Technologies (ICSIT07).

本研究旨在透过语音辨识与文字显示，为年龄相关的失聪人士开发对话支援系统。为此，预计该系统将体积小、重量轻。在研究之初，我们打算将自定义ISI与微显示相结合来设计系统。然而，由于内置麦克风的移动计算机最近进入市场，我们改变了在移动计算机上实现系统。2006年初，该系统的设计初衷是只识别预先注册的单词，因此未注册的单词无法识别就显得很不方便。因此，我们改进了系统，使其能够识别所有的日语101个音节（假名）。在这种情况下，对话伙伴必须一个音节一个音节地说话，系统将识别出来的音节一个接一个地显示出来。此外，系统以一串假名字符来显示用餐内容也带来了不便，听觉上难以阅读。2007年，我们从以下三点完善了这一制度：首先，结合我们提出的新算法提取噪声鲁棒性语音特征，降低识别误差。二是在系统中加入“假名转汉字”功能，该功能的命令也是通过语音来命令的；三是使系统显示识别的上升候选项，在第一个候选项识别错误时可以选择第二个候选项。这个选择命令也可以通过语音来接受。通过使系统可以仅通过语音命令进行操作，使系统使用起来更加方便。在识别速度和识别误差方面有待进一步提高。部分研究结果已在日本声学学会演讲技术委员会、电子、信号和通信工程师学会以及通信和信息技术国际研讨会（ICSIT07）上发表。

项目成果

A consideration on extracting temporal feature of Spectrum for speech recognition

语音识别中频谱时间特征提取的思考

• 发表时间: 2007
• 作者: Megumi;Umeno
• 通讯作者: Umeno

短時間パワースペクトルの時間・周波数傾斜の3値化に基づく音声特徴量を用いた音声認識

使用基于短期功率谱的时间和频率斜率三值化的语音特征的语音识别

• 发表时间: 2006
• 期刊: 電子情報通信学会音声研究会技術研究報告 SP2006-31
• 作者: Tohgoroh Matsui;Hayato Ohwada;Kazuyuki Kuchitsu.;池田晃;山崎敦;影山勝彦;冨樫由美子;Shinichiro Omachi;岩村雅一;内田誠一;大町真一郎;Kazuya Negishi;Katsuhiko Kageyama;根岸和也;Koichi Kise;Shinichiro Omachi;Seiichi Uchida;大町真一郎;堀内 真;岩村雅一;古谷嘉男;古谷嘉男;田中一弘;Megumi Sakai;古谷嘉男;岩村雅一;Yoshio Furuya;梅野 恵;Masakazu Iwamura;黒田 弥文;古谷嘉男;梅野 恵
• 通讯作者: 梅野 恵

音声認識のためのスペクトルの時間変化を表す特徴量抽出についての検討

语音识别中表示频谱时间变化的特征提取研究

• 发表时间: 2007
• 作者: 木下浩二;村上研二;Seiichi Uchida;梅野 恵
• 通讯作者: 梅野 恵

スペクトル傾斜の3値化に基づく特徴量を用いた雑音下単語認識システム

基于谱斜率三值化特征的噪声词识别系统

• 发表时间: 2007
• 期刊: 日本音響学会研究発表会講演論文集 2007春季(CD-ROM)
• 作者: Tohgoroh Matsui;Hayato Ohwada;Kazuyuki Kuchitsu.;池田晃;山崎敦;影山勝彦;冨樫由美子;Shinichiro Omachi;岩村雅一;内田誠一;大町真一郎;Kazuya Negishi;Katsuhiko Kageyama;根岸和也;Koichi Kise;Shinichiro Omachi;Seiichi Uchida;大町真一郎;堀内 真;岩村雅一;古谷嘉男;古谷嘉男;田中一弘;Megumi Sakai;古谷嘉男;岩村雅一;Yoshio Furuya;梅野 恵
• 通讯作者: 梅野 恵

On noise robustness of feature expressing temporal variation for word speech recognition

词语语音识别中表达时间变化特征的噪声鲁棒性

• 发表时间: 2007
• 作者: Tetsuo;Funada
• 通讯作者: Funada