专利快速检索-快速检索全球专利，免费商用专利数据库-IPRDB

51. 发明申请

WO03001508B1 METHOD FOR MULTIPLE ACCESS AND TRANSMISSION IN A POINT-TO-MULTIPOINT SYSTEM ON AN ELECTRIC NETWORK 审中-公开
标题翻译：一种电网络中点对多点系统中多种接入和传输的方法
公开(公告)号：WO03001508B1
公开(公告)日：2004-07-08
申请号：PCT/ES0200312
申请日：2002-06-25
申请人： UNI POMPEU FABRA , BATLLE I MONT ELOI
发明人： BATLLE I MONT ELOI
IPC分类号： G10H1/00 , G10L15/14 , G10L19/26 , G10L101/027 , G10L101/065 , G10L101/10
CPC分类号： G10L19/265 , G10H1/00 , G10H2250/015 , G10H2250/021 , G10L15/142
摘要： Said method comprises the following stages: 1. Preprocessing (602) the audio sequence, which involves the steps of eliminating frequencies above a predetermined value with a low-pass filter and digitizing the signal in an analog/digital converter; 2. Extracting parameters(301 ) representing the audio sequence with the purpose of obtaining a parameter vector Ot especially adapted to the intended identification approach; 3. Computing abstract descriptors (302) representing parameter vector Ot, which have been implemented as Hidden Markov Models optimized by the use of a database defining abstract descriptors (303), which has been generated during prior execution of a first phase in method learning mode; 4. Identifying (605) audio sequences thus treated in an abstract descriptor sequence database (505) generated during prior execution of a second phase in method learning mode; 5. Recording the results (607) obtained in the identification stage (605).
摘要翻译：所述方法包括以下步骤：1.预处理（602）音频序列，该步骤包括以下步骤：用低通滤波器消除高于预定值的频率，并在模拟/数字转换器中数字化该信号; 2.提取表示音频序列的参数（301），目的是获得特别适用于预期识别方法的参数向量Ot; 3.计算表示参数向量Ot的抽象描述符（302），其已经实施为利用定义抽象描述符（303）的数据库优化的隐马尔可夫模型，所述抽象描述符在方法学习模式中的第一阶段的先前执行期间 ; 4.在方法学习模式中识别（605）在先前执行第二阶段期间生成的抽象描述符序列数据库（505）中如此处理的音频序列; 5.记录在识别阶段（605）中获得的结果（607）。

52. 发明申请

WO2004036546A1 CLASSIFICATION OF VECTORS IN NOISY CONDITIONS 审中-公开
标题翻译：噪声条件下的矢量分类
公开(公告)号：WO2004036546A1
公开(公告)日：2004-04-29
申请号：PCT/GB2003/004111
申请日：2003-09-15
申请人： THE QUEEN'S UNIVERSITY OF BELFAST , MING, Ji
发明人： MING, Ji
IPC分类号： G10L15/20
CPC分类号： G06K9/6228 , G10L15/142 , G10L15/20
摘要： A method for vector classification employs a statistical method, the posterior union model, for signal processing and pattern classification in noisy conditions, requiring no knowledge about the noise characteristics. According to the method, avector X to be classified comprises N components, M of which are corrupt, and belongs to one of Q classes C 1 , C 2 , ..., C Q , and where X N-M denotes the subset of the (N-M) clean components in vector X, the method comprising the step of; performing a maximum a posteriori (MAP) probability decision to determine the class C i to which the vector X is deemed to belong; wherein the largest modulus value of P (C i | X N-M ) , is determined for specific values of M and for specific C i ' s, and wherein P (X N-M | C i ) is calculated by performing a disjunction over all possible subsets of (N-M) components taken from X.
摘要翻译：用于矢量分类的方法采用统计方法，后联合模型，用于噪声条件下的信号处理和模式分类，不需要关于噪声特性的知识。根据该方法，待分类的矢量X包括N个分量，其中M个被破坏，并且属于Q个类C1，C2，...，CQ中的一个，并且其中，XN-M表示（NM）在向量X中的清洁组件，该方法包括以下步骤：执行最大后验（MAP）概率决定来确定矢量X被认为属于的类Ci; 其中对于M的特定值和对于特定Ci的确定P（Ci || XN-M）的最大模数值，并且其中P（XN-M || Ci）是通过执行所有可能的分离来计算的（NM）组分的子集取自X.

53. 发明申请

WO2004010413A1 METHOD AND APPARATUS FOR DIFFERENTIAL COMPRESSION OF SPEAKER MODELS FOR SPEAKER RECOGNITION 审中-公开
标题翻译：用于语音识别的扬声器模型的差分压缩的方法和装置
公开(公告)号：WO2004010413A1
公开(公告)日：2004-01-29
申请号：PCT/US2003/016350
申请日：2003-05-23
申请人： MASSACHUSETTS INSTITUTE OF TECHNOLOGY , REYNOLDS, Douglas, A.
发明人： REYNOLDS, Douglas, A.
IPC分类号： G10L17/00
CPC分类号： G10L15/07 , G10L15/142 , G10L15/285 , Y10S707/99942
摘要： A differential compression technique is disclosed for compression individual speaker models, such as Gaussian mixture models, by computing a delta model from the difference between an individual speaker model and a baseline model. Further compression may be applied to the delta model to reduce the large storage requirements generally attributed to speaker models.
摘要翻译：公开了一种差分压缩技术，用于压缩单个扬声器模型，例如高斯混合模型，通过根据个人扬声器模型和基准模型之间的差异计算增量模型。进一步的压缩可以应用于delta模型，以减少通常归因于说话者模型的大的存储要求。

54. 发明申请

WO2003083832A1 PROCÉDÉ DE RECONNAISSANCE DE PAROLE AU MOYEN D'UN TRANSDUCTEUR UNIQUE 审中-公开
标题翻译：使用单个传感器的语音识别方法
公开(公告)号：WO2003083832A1
公开(公告)日：2003-10-09
申请号：PCT/FR2003/000884
申请日：2003-03-20
申请人： FRANCE TELECOM SA , FERRIEUX, Alexandre , DELPHIN-POULAT, Lionel
发明人： FERRIEUX, Alexandre , DELPHIN-POULAT, Lionel
IPC分类号： G10L15/28
CPC分类号： G10L15/183 , G10L15/142 , G10L15/285 , G10L2015/025
摘要： La présente invention concerne un procédé de traduction de données d'entrée AVin en une séquence lexicale de sortie OUTSQ, au cours duquel des entités souslexicales et diverses combinaisons possibles desdites entités sont identifiées en tant qu'états ei et ej de premier et deuxième modèles de langage APMM et PHLM, respectivement, destinés à être mémorisés, avec une valeur de vraisemblance Sij associée, dans une table TAB munie de zones mémoire dont chacune est destinée à contenir au moins une combinaison d'états (ei;ej) et est munie d'une adresse égale à une valeur h[(ei;ej)] d'une fonction scalaire h appliquée à des paramètres propres à la combinaison (ei;ej). L'invention permet de limiter la complexité d'accès aux informations produites par un transducteur unique formé par une unique machine de Viterbi VM3 exploitant lesdits modèles APMM et PHLM.
摘要翻译：本发明涉及将输入数据AVin转换为输出词汇序列OUTSQ的方法。在所述方法期间，所述实体的子词汇实体和不同的可能组合分别被识别为第一和第二语言模型APMM和PHLM的状态ei和ej。所述组合旨在与包括存储区域的表TAB中的关联似然值Sij一起存储。此外，每个所述存储区域旨在包含状态（ei; ej）的至少一种组合，并且被提供有等于应用于参数的标量函数h的值h [（ei; ej）]的地址特定于组合（ei; ej）。本发明可以用于限制由单个维特比VM3机器操作模型APMM和PHLM形成的单个换能器产生的访问信息的复杂性。

55. 发明申请

WO02007146A1 SPEECH RECOGNITION DEVICE AND SPEECH RECOGNITION METHOD 审中-公开
标题翻译：语音识别装置和语音识别方法
公开(公告)号：WO02007146A1
公开(公告)日：2002-01-24
申请号：PCT/JP2001/006092
申请日：2001-07-13
IPC分类号： G10L15/14 , G10L15/28
CPC分类号： G10L15/32 , G10L15/142 , G10L15/28
摘要： Each word to be recognized is represented by hidden Markov models for male and female and an output probability function and a transition probability preset in hidden Markov models for male and female are prestored in a ROM (6). With reference to feature parameters detected by a feature detecting section (3) and the hidden Markov models, a speech recognizing section (4) determines an occurrence probability of a feature parameter sequence. In the process for determining the occurrence probability, the speech recognizing section (4) gives each word one state sequence of a hidden Markov model common to the hidden Markov models for male and female, multiplies an output probability function value by a transition probability of a preset combination among the output probability functions and transition probabilities stored in the ROM (6), selects a maximum product, determines the occurrence probability based on the selected product, and then recognizes the input speech based on the occurrence probability thus determined.
摘要翻译：要识别的每个单词由男性和女性的隐马尔可夫模型表示，男性和女性的隐马尔可夫模型中预测的输出概率函数和转移概率预先存储在ROM中（6）。参考由特征检测部分（3）检测的特征参数和隐马尔可夫模型，语音识别部分（4）确定特征参数序列的出现概率。在确定发生概率的过程中，语音识别部分（4）给出男性和女性的隐马尔科夫模型共同的隐马尔可夫模型的每个单词一个状态序列，将输出概率函数值乘以一个存储在ROM（6）中的输出概率函数和转移概率之间的预设组合选择最大乘积，基于所选择的乘积确定出现概率，然后基于所确定的发生概率来识别输入语音。

56. 发明申请

WO01056020A1 METHOD AND DEVICE FOR CREATING A TEXT FILE BY USING VOICE RECOGNITION 审中-公开
标题翻译：方法和设备，用于创建一个文本文件，通过语音识别
公开(公告)号：WO01056020A1
公开(公告)日：2001-08-02
申请号：PCT/DE2001/000052
申请日：2001-01-09
IPC分类号： G10L15/06 , G10L15/14 , G10L15/22 , G10L15/30 , H04M1/27 , H04M1/725 , H04M3/53 , H04M3/533 , G10L15/26
CPC分类号： G10L15/30 , G10L15/063 , G10L15/142 , G10L2015/0631 , H04M1/271 , H04M1/72547 , H04M3/5322 , H04M3/53316 , H04M2201/40 , H04M2201/60 , H04M2207/18
摘要： According to the invention, a mobile telephone is provided as a terminal for voice input, and the mobile telephone is designed for preprocessing the voice input in the form of a voice analysis with the output of digital feature vectors. Devices are provided for transmitting the preprocessed signal via the mobile telephone to a computer node. The computer node has a voice recognizer for processing the preprocessed signal, and a device for returning an output of the HMM recognizer as digital text to the mobile telephone is provided in the computer node.
摘要翻译：本发明的特征在于，移动电话被作为用于语音输入的终端提供，移动电话用于预处理所述语音输入是在语音分析，以数字特征向量的输出的形式，提供了用于将所述预处理信号的传输通过手机到计算机节点中，计算机节点具有用于处理该预处理的信号的语音识别器，并在计算机节点包括：用于向移动电话的存在返回的HMM识别器的输出作为数字文本。

57. 发明申请

WO01035391A1 DISTRIBUTED REAL TIME SPEECH RECOGNITION SYSTEM 审中-公开
标题翻译：分布式实时语音识别系统
公开(公告)号：WO01035391A1
公开(公告)日：2001-05-17
申请号：PCT/US2000/030918
申请日：2000-11-10
IPC分类号： G06F3/16 , G06F17/28 , G06F17/30 , G09B7/02 , G10L13/00 , G10L15/00 , G10L15/02 , G10L15/14 , G10L15/18 , G10L15/20 , G10L15/22 , G10L15/28 , G10L21/02
CPC分类号： G06F17/3043 , G10L15/142 , G10L15/1815 , G10L15/183 , G10L15/285 , G10L15/30 , H04M2250/74
摘要： A real-time system (100) incorporating speech recognition and linguistic processing for recognizing a spoken query by a user and distributed between client (150) and server (180), is disclosed. The system (100) accepts user's queries in the form of speech at the client (150) where minimal processing extracts a sufficient number of acoustic speech vectors representing the utterance. These vectors are sent via a communications channel (160A) to the server (180) where additional acoustic vectors are derived. Using Hidden Markov Models (HMMs), and appropriate grammars and dictionaries conditioned by the selections made by the user, the speech representing the user's query is fully decoded into text (or some other suitable form) at the server (180). The text corresponding to the user's query is then simultaneously sent to a natural language engine (190) and a database processor (186) where optimized SQL statements are constructed for a full-text search from a database (188) for a record set of several stored questions that best matches the user's query. Further processing in the natural language engine (190) narrows the search to a single stored question. The answer corresponding to this single stored question is next retrieved from the file path and sent to the client (150) in compressed form. At the client (150), the answer to the user's query is articulated to the user using a text-to-speech engine (159) in his or her native natural language. The system (100) requires no training and can operate in several natural languages.
摘要翻译：公开了一种结合语音识别和语言处理的实时系统（100），用于识别由用户进行的口语查询并且分布在客户端（150）与服务器（180）之间。系统（100）以客户端（150）的语音形式接受用户的查询，其中最小处理提取足够数量的表示话语的声学语音向量。这些向量经由通信信道（160A）发送到服务器（180），其中导出附加的声学矢量。使用隐马尔可夫模型（HMM）以及由用户做出的选择所适用的适当的语法和词典，表示用户查询的语音在服务器（180）被完全解码成文本（或某种其他合适的形式）。然后将与用户查询相对应的文本同时发送到自然语言引擎（190）和数据库处理器（186），其中针对数据库（188）为数据库（188）构建用于全文搜索的优化SQL语句，用于多个记录集存储与用户查询最匹配的问题。自然语言引擎（190）中的进一步处理将搜索缩小为单个存储的问题。对应于该单个存储的问题的答案接下来从文件路径检索并以压缩形式发送给客户端（150）。在客户端（150），使用他或她的母语自然语言的文本到语音引擎（159）将用户查询的答案表达给用户。系统（100）不需要训练，可以使用多种自然语言进行操作。

58. 发明申请

WO98043238A1 SPEECH ANALYSIS SYSTEM 审中-公开
标题翻译：语音分析系统
公开(公告)号：WO98043238A1
公开(公告)日：1998-10-01
申请号：PCT/GB1998/000615
申请日：1998-02-26
IPC分类号： G10L15/02 , G10L15/065 , G10L15/14 , G10L15/20 , H04M1/00 , G10L3/00
CPC分类号： G10L15/065 , G10L15/142 , G10L15/20
摘要： A speech analysis system (10) incorporates a filterbank analyser (18) producing successive frequency data vectors for a speech signal from two speakers. From each data vector, units (22A and 22B) produce a set of modified data vectors compensated for differing forms of distortion associated with respective speakers. A computer (24) matches modified data vectors to hidden Markov model states. It identifies the modified data vector in each set exhibiting greatest matching probability, the model state matched therewith, the form of distortion with which it is associated and the model class, i.e. speech or noise. The matched model state has a mean value providing an estimate of its associated data vector. The estimate is compared with its associated data vector, and their difference is averaged with others associated with a like form of distortion in an infinite response filter bank (48A or 48B) to provide compensation for that form of distortion. Averaged difference vectors provide compensation for multiple forms of distortion associated with respective speakers.
摘要翻译：语音分析系统（10）包括滤波器组分析器（18），其产生用于来自两个扬声器的语音信号的连续的频率数据矢量。从每个数据向量，单元（22A和22B）产生一组经修正的数据向量，这些修改的数据向量被补偿，用于与各个扬声器相关联的不同形式的失真计算机（24）将修改的数据向量与隐马尔可夫模型状态相匹配。它识别表现出最大匹配概率的每个集合中的修改的数据向量，与其匹配的模型状态，与其相关联的失真的形式和模型类，即语音或噪声。匹配模型状态具有提供其关联数据向量的估计的平均值。将估计与其相关联的数据向量进行比较，并且将它们的差异与在无限响应滤波器组（48A或48B）中与相似形式的失真相关联的其他差异进行平均，以提供对该形式的失真的补偿。平均差矢量为与各个扬声器相关联的多种形式的失真提供补偿。

59. 发明申请

WO9810412A3 SPEECH RECOGNITION AND VERIFICATION SYSTEM ENABLING AUTHORIZED DATA TRANSMISSION OVER NETWORKED COMPUTER SYSTEMS 审中-公开
标题翻译：语音识别和验证系统通过网络计算机系统启用授权数据传输
公开(公告)号：WO9810412A3
公开(公告)日：1998-07-09
申请号：PCT/US9715943
申请日：1997-09-09
申请人： VOICE CONTROL SYSTEMS INC
发明人： WEIDEMAN WILLIAM E
IPC分类号： G07C9/00 , G10L15/02 , G10L15/14 , G10L17/00 , G10L
CPC分类号： G06Q20/425 , G07C9/00158 , G10L15/142 , G10L17/24
摘要： A system and apparatus for using speech recognition and verification to provide secure and authorized data transmissions between networked computer systems is provided. The system includes first and second network computer systems wherein a request for a transaction by user of the first computer system causes the user to be prompted to enter a spoken identifier such as a credit card number, PIN number or password. This spoken identifier is converted from speech data into speech feature data using either a resident software application or a downloaded application from the second computer system. The speech feature data is transmitted to the second computer system wherein speech recognition and verification engines identify the spoken identifier and determine whether or not the user who spoke the identifier is properly associated with the spoken identifier. Upon successful completion of this recognition and verification process, the requested transaction is completed.

60. 发明申请

WO1998022937A1 SPEECH PROCESSING SYSTEM 审中-公开
标题翻译：语音处理系统
公开(公告)号：WO1998022937A1
公开(公告)日：1998-05-28
申请号：PCT/GB1997002818
申请日：1997-10-13
申请人： THE SECRETARY OF STATE FOR DEFENCE , HOLMES, John, Nicholas
发明人： THE SECRETARY OF STATE FOR DEFENCE
IPC分类号： G10L09/06
CPC分类号： G10L15/10 , G10L15/12 , G10L15/142 , G10L25/15
摘要： A speech processing system (10) incorporates an analogue to digital converter (16) to digitise input speech signals for Fourier transformation to produce short-term spectral cross-sections. These cross-sections are compared with one hundred and fifty reference patterns in a store (34), the patterns having respective stored sets of formant frequencies assigned thereto by a human expert. Six stored patterns most closely matching each input cross-section are selected for further processing by dynamic programming, which indicates the pattern which is a best match to the input cross-section by using frequency-scale warping to achieve alignment. The stores formant frequencies of the best matching pattern are modified by the frequency warping, and the results are used as formant frequency estimates for the input cross-section. The frequencies are further refined on the basis of the shape of the input cross-section near to the chosen formants. Formant amplitudes are produced from input cross-section amplitudes at estimated formant frequencies. The formant frequencies and amplitudes are used with a computer (25) to provide speech indications or with a Hidden Markov Model word matcher (24) to provide word recognition.
摘要翻译：语音处理系统（10）包括模数转换器（16），用于数字化用于傅立叶变换的输入语音信号以产生短期频谱横截面。将这些横截面与商店（34）中的一百五十个参考图案进行比较，这些图案具有由人类专家分配给它们的各自存储的共振峰频率集合。选择最接近匹配每个输入横截面的六个存储模式，以通过动态规划进一步处理，其通过使用频率尺度翘曲来指示与输入横截面最佳匹配的模式以实现对准。通过频率扭曲修改最佳匹配模式的存储共振峰频率，并将结果用作输入截面的共振峰频率估计。基于所选择的共振峰附近的输入横截面的形状，进一步改进频率。在预计的共振峰频率下，从输入横截面振幅产生共振幅度。共振峰频率和幅度与计算机（25）一起使用以提供语音指示或与隐马尔可夫模型词匹配器（24）提供字识别。

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