专利快速检索-快速检索全球专利，免费商用专利数据库-IPRDB

91. 发明专利

JP2008191551A Pronunciation evaluating apparatus and program 有权
标题翻译：授权评估设备和程序
公开(公告)号：JP2008191551A
公开(公告)日：2008-08-21
申请号：JP2007027903
申请日：2007-02-07
申请人： Advanced Telecommunication Research Institute International , 株式会社国際電気通信基礎技術研究所
发明人： TAGAWA HIROAKI , ADACHI TAKAHIRO , WATANABE HIDEYUKI , KUBO RIEKO , KOMAKI AKIRA , IKUMA HIROKO , YAMADA REIKO
IPC分类号： G10L15/00 , G09B19/00 , G09B19/04 , G10L11/00 , G10L15/10 , G10L15/14
摘要： PROBLEM TO BE SOLVED: To solve the following problem: evaluation accuracy by phoneme units is not sufficient. SOLUTION: A pronunciation evaluating apparatus obtains a feature vector series from a teacher data and one or more frame voice data, compares the feature vector series with an acoustic model according to a phoneme series to be evaluated, obtains an optimal state series which is a set of an optimal state for each frame, identifies one or more optimal phoneme series in which the same phoneme continues among the optimal state series, obtains one or more optimal phoneme portion series which are groups of the one or more optimal phoneme series, obtains one or more feature vector portion series which are groups of one or more feature vectors corresponding to each optimal phoneme portion series. obtains a posterior probability at which the feature vector series is a phoneme to be evaluated, and calculates an evaluation value from the posterior probability. The evaluation by phoneme units is highly accurately performed by the pronunciation evaluating apparatus. COPYRIGHT: (C)2008,JPO&INPIT
摘要翻译：要解决的问题：解决以下问题：音素单位的评估精度是不够的。解答：发音评价装置从教师数据和一个或多个帧语音数据中获取特征向量序列，根据要评估的音素系列将特征向量序列与声学模型进行比较，获得最佳状态序列，其中是对于每个帧的一组最优状态，识别在最优状态序列中相同音素继续的一个或多个最优音素系列，获得作为一个或多个最佳音素系列的组的一个或多个最优音素部分序列，获得一个或多个特征向量部分序列，其是与每个最佳音素部分序列相对应的一个或多个特征向量的组。获得特征向量序列是要评估的音素的后验概率，并根据后验概率计算评价值。通过发音评价装置高精度地进行音素单元的评价。版权所有（C）2008，JPO＆INPIT

92. 发明专利

JP2008513825A Robust speech recognition system that does not depend on the speaker 有权
公开(公告)号：JP2008513825A
公开(公告)日：2008-05-01
申请号：JP2007531910
申请日：2005-09-13
申请人：コーニンクレッカフィリップスエレクトロニクスエヌヴィ
发明人：ディーテルゲレル
IPC分类号： G10L15/06 , G10L15/07 , G10L15/14
CPC分类号： G10L15/07 , G10L15/063 , G10L15/144 , G10L15/20
摘要：本発明は、複数の環境条件及び複数の話者に対するトレーニングデータを提供する話者非依存の音声認識システムに話者依存表現を組み込む方法を提供する。話者依存表現は、特徴ベクトルのシーケンスに変換され、生成された特徴ベクトルのシーケンスに対する最小距離を持つ話者非依存のトレーニングデータのセットの混合確率密度が決定される。そして決定された混合確率密度が、話者依存表現の隠れマルコフモデル(HMM)の状態に割り当てられる。従って、もはや話者依存のトレーニングデータ及び基準が、音声認識システムに明示的に格納されることはない。更に、話者依存表現を話者非依存のトレーニングデータで表すことにより、本質的に環境適合が提供される。更に、本発明は、話者依存表現に基づき、変化する環境条件に対する音声認識システムの堅牢性を実質的に改善する人工的な特徴ベクトルの生成を提供する。

93. 发明专利

JP4061094B2 Voice recognition device, the speech recognition method and program 有权
公开(公告)号：JP4061094B2
公开(公告)日：2008-03-12
申请号：JP2002072456
申请日：2002-03-15
申请人：インターナショナル・ビジネス・マシーンズ・コーポレーションInternational Business Maschines Corporation
发明人：哲也滝口 , 雅史西村
IPC分类号： G10L15/14 , G10L15/02 , G10L15/06 , G10L15/08 , G10L15/12 , G10L15/20 , G10L15/28
CPC分类号： G10L15/30 , G10L15/02 , G10L15/20
摘要： Speech recognition is performed by matching between a characteristic quantity of an inputted speech and a composite HMM obtained by synthesizing a speech HMM (hidden Markov model) and a noise HMM for each speech frame of the inputted speech by use of the composite HMM.

94. 发明专利

JP4058521B2 Correction processing method of background noise distortion and speech recognition system using the same 有权
公开(公告)号：JP4058521B2
公开(公告)日：2008-03-12
申请号：JP2003320183
申请日：2003-09-11
申请人：独立行政法人産業技術総合研究所
发明人：晃佐宗
IPC分类号： G10L15/06 , G10L15/10 , G10L15/14
摘要： PROBLEM TO BE SOLVED: To provide a correction processing method for background noise distortion that can improve correction precision even when a noise signal changes and a speech recognition system using the same. SOLUTION: A speech signal including a noise signal inputted from a microphone 1 (having an A/D converter inside) is inputted to a speech information processing part 2. This speech information processing part 2 is equipped with a section (frame) detection part 3 based upon the power of the input signal, a calculation part 4 for an input feature quantity time series, a calculation part 5 for cumulative output probability of HMMs calculated by Viterbi algorithm, a postprobability calculation part 6 for a distribution of HMMs which divides an input feature quantity time series and the cumulative output probability of HMMs by the number of frames and performs weighting with a value converted into an antilog value, a distortion suppression part 7 for a speech feature quantity including the noise signal, and a speech recognition part 8. COPYRIGHT: (C)2005,JPO&NCIPI

95. 发明专利

JP2008015120A Voice recognition apparatus and its method 有权
标题翻译：语音识别装置及其方法
公开(公告)号：JP2008015120A
公开(公告)日：2008-01-24
申请号：JP2006185002
申请日：2006-07-04
申请人： Toshiba Corp , 株式会社東芝
发明人： SAKAI MASARU , TANAKA SHINICHI
IPC分类号： G10L15/08 , G10L15/14
CPC分类号： G10L15/142 , G10L2015/085
摘要： PROBLEM TO BE SOLVED: To provide a voice recognition apparatus capable of efficiently reducing the number of cycles of output probability calculation without negatively affecting voice recognition performance by combining a method based on beam search and a method based on a reference frame. SOLUTION: A sound processing section 101, a voice interval detection section 102, a dictionary section 103, a collating section 104, a search object selection section 105, a storing section 106 and a determination section 107, are provided. Processing comprises the steps of: selecting a search range based on the beam search; setting and storing the reference frame; storing the output probability in a transition path; and determining whether the output probability in the transition path is stored or not. The search range is selected based on the beam search, and the output probability of the transition path is calculated only once for an interval from setting time to updating time of the reference frame, and a value of the calculation is stored, and when the output probability of the transition path is stored in the following frames, the stored value is set as an approximation value of the output probability. Thereby, the frequency of the output probability calculation is reduced. COPYRIGHT: (C)2008,JPO&INPIT
摘要翻译：要解决的问题：提供一种能够通过组合基于波束搜索的方法和基于参考帧的方法来有效地减少输出概率计算的周期数而不会不利地影响语音识别性能的语音识别装置。解决方案：提供声音处理部分101，语音间隔检测部分102，字典部分103，对照部分104，搜索对象选择部分105，存储部分106和确定部分107。处理包括以下步骤：基于波束搜索选择搜索范围; 设置和存储参考系; 将输出概率存储在转换路径中; 并确定过渡路径中的输出概率是否被存储。基于波束搜索选择搜索范围，并且从参考帧的设置时间到更新时间的间隔仅计算一次转换路径的输出概率，并存储计算值，并且当输出将转换路径的概率存储在随后的帧中，将存储的值设置为输出概率的近似值。从而减少输出概率计算的频率。版权所有（C）2008，JPO＆INPIT

96. 发明专利

JPWO2005048239A1 音声認識装置有权
公开(公告)号：JPWO2005048239A1
公开(公告)日：2007-11-29
申请号：JP2005515466
申请日：2004-11-12
申请人：本田技研工業株式会社
发明人：一博中臺 , 一博中臺 , 辻野　広司 , 広司辻野 , 博奥乃 , 博奥乃 , 山本　俊一 , 俊一山本
IPC分类号： G10L15/06 , G10L15/14 , G10L15/20 , G10L15/28 , G10L21/02
CPC分类号： G10L15/20 , G10L21/028 , G10L2021/02166
摘要：話者や、音声認識装置を搭載した移動体が移動しても高い精度で認識可能な音声認識装置を提供する。複数のマイクMが検出した音響信号から、特定の話者HMjの音声を認識して文字情報に変換する音声認識装置である。複数のマイクMが検出した音響信号に基づき、話者HMjの音源方向θHMjを特定する音源定位部10と、音源方向θHMjに基づき、話者HMjの音声信号を音響信号から分離する音源分離部20と、断続的な複数の方向に対応した方向依存音響モデルH(θn)を記憶した音響モデル記憶部49と、音源方向θHMjの音響モデルを、音響モデル記憶部49の方向依存音響モデルH(θn)に基づいて求め、音響モデル記憶部49へ記憶させる音響モデル合成部40と、音響モデル合成部40が合成した音響モデルを使用して、音源分離部20が分離した音声信号の音声認識を行い、文字情報に変換する音声認識部50とを備える。

97. 发明专利

JP4004368B2 Voice-recognition system 有权
公开(公告)号：JP4004368B2
公开(公告)日：2007-11-07
申请号：JP2002277541
申请日：2002-09-24
申请人：株式会社国際電気通信基礎技術研究所
发明人：ヤオ・カイシェン , 哲中村 , 建武党
IPC分类号： G10L15/14

98. 发明专利

JP2007279349A Feature amount compensation apparatus, method, and program 有权
标题翻译：特征补偿装置，方法和程序
公开(公告)号：JP2007279349A
公开(公告)日：2007-10-25
申请号：JP2006105091
申请日：2006-04-06
申请人： Toshiba Corp , 株式会社東芝
发明人： AKAMINE MASAMI , MASUKO TAKASHI , BARREDA DANIEL , REMCO TEUNEN
IPC分类号： G10L15/20 , G10L15/14
CPC分类号： G10L15/20 , G10L15/02 , G10L15/065
摘要： PROBLEM TO BE SOLVED: To provide a feature amount compensation apparatus capable of calculating a feature amount with high accuracy.
SOLUTION: The feature amount compensation apparatus comprises: a noise environment storage section 120 for storing a first compensation amount for each of a plurality of noise environments; a feature extraction section 102 for extracting a feature amount of input voice; a reversion calculation section 103 for calculating reversion of input voice for each noise environment, based on the feature amount; a compensation vector calculation section 104 for calculating a second compensation value, based on the first compensation value obtained from the noise environment storage section 120, and for calculating a third compensation value in which the calculated second compensation value is weighted and added with the reversion as a weighting coefficient; and a feature vector compensation section 105 for compensating the feature amount extracted, based on the third compensation amount.
COPYRIGHT: (C)2008,JPO&INPIT
摘要翻译：要解决的问题：提供能够以高精度计算特征量的特征量补偿装置。解决方案：特征量补偿装置包括：噪声环境存储部分120，用于存储多个噪声环境中的每一个的第一补偿量; 用于提取输入声音的特征量的特征提取部分102; 反转计算部分103，用于基于特征量计算每个噪声环境的输入声音的反转; 补偿矢量计算部分104，用于基于从噪声环境存储部分120获得的第一补偿值来计算第二补偿值，并且用于计算将所计算的第二补偿值加权并加上逆转的第三补偿值，加权系数; 以及用于基于第三补偿量补偿所提取的特征量的特征向量补偿部分105。版权所有（C）2008，JPO＆INPIT

99. 发明专利

JP2007240552A Musical instrument sound recognition method, musical instrument annotation method and music piece searching method 审中-公开
标题翻译：音乐仪器声音识别方法，音乐仪器声调方法和音乐片段搜索方法
公开(公告)号：JP2007240552A
公开(公告)日：2007-09-20
申请号：JP2006058649
申请日：2006-03-03
申请人： Kyoto Univ , 国立大学法人京都大学
发明人： KITAHARA TETSURO , OKUNO HIROSHI
IPC分类号： G10L11/00 , G10G3/04 , G10L15/10 , G10L15/14 , G10L21/06
摘要： PROBLEM TO BE SOLVED: To provide a new musical instrument sound recognition method which is not dependent on sound output time and F0 estimation processing. SOLUTION: The musical instrument sound recognition method comprises the steps of: calculating an unspecified musical instrument existence probability; calculating a conditional musical instrument existence probability; and calculating a product of both probabilities. In the musical instrument sound recognition method, recognition accuracy is higher than a conventional method, since the musical existence probability is calculated without performing musical sound recognition for each single tone. Moreover, it is easy to visualize it as Instrogram on a time t - frequency f plane, and by using this, various applications can be considered, such as, performing musical instrument annotation and performing similar music searching based on a music piece structure. COPYRIGHT: (C)2007,JPO&INPIT
摘要翻译：要解决的问题：提供不依赖于声音输出时间和F0估计处理的新乐器声音识别方法。乐器声音识别方法包括以下步骤：计算未指定的乐器存在概率; 计算条件乐器存在概率; 并计算两种概率的乘积。在乐器声音识别方法中，识别精度高于常规方法，因为在不对每个单音进行音乐声音识别的情况下计算出音乐存在概率。此外，很容易将其可视化为时间t频率f平面上的Instrogram，并且通过使用它，可以考虑各种应用，诸如执行乐器注释并且基于音乐片结构执行类似的音乐搜索。版权所有（C）2007，JPO＆INPIT

100. 发明专利

JP3965141B2 Speech recognition device 有权
公开(公告)号：JP3965141B2
公开(公告)日：2007-08-29
申请号：JP2003293836
申请日：2003-08-15
申请人：株式会社国際電気通信基礎技術研究所
发明人：哲中村 , 俊樹遠藤
IPC分类号： G10L15/00 , G10L15/20 , G10L15/14 , G10L15/28 , G10L19/00 , G10L21/02
摘要： PROBLEM TO BE SOLVED: To provide a speech recognition apparatus for relatively highly exactly recognizing speech even when large loss is present in a sequence of feature parameters of a speech. SOLUTION: The speech recognition apparatus includes an input buffer 30 which receives a speech data frame, a frame buffer 36 which stores frames in the order of frame numbers, a frame loss detection part 32 which detects the occurrence of frame loss, a feature parameter estimation part 34 which estimates feature data of frames as many as loss frames on the basis of feature data in the frame buffer 36 and frame order information and inserts the estimated feature data into the prescribed position in the frame buffer 36, and a speech recognition part 38 which reads frames out of the feature parameter estimation part 34 in order to perform speech recognition. COPYRIGHT: (C)2005,JPO&NCIPI

你已经成功收藏专利！

检索式保存成功!

IPRDB

热门服务

关于我们

友情链接

联系方式