公开(公告)号：US07933766B2

公开(公告)日：2011-04-26

申请号：US12582062

申请日：2009-10-20

申请人： Narendra K. Gupta ,  Mazin G. Rahim ,  Gokhan Tur ,  Antony Van der Mude

发明人： Narendra K. Gupta ,  Mazin G. Rahim ,  Gokhan Tur ,  Antony Van der Mude

IPC分类号： G06F17/27

CPC分类号： G10L15/193 ,  G10L15/063 ,  G10L15/183

摘要： A method of generating a natural language model for use in a spoken dialog system is disclosed. The method comprises using sample utterances and creating a number of hand crafted rules for each call-type defined in a labeling guide. A first NLU model is generated and tested using the hand crafted rules and sample utterances. A second NLU model is built using the sample utterances as new training data and using the hand crafted rules. The second NLU model is tested for performance using a first batch of labeled data. A series of NLU models are built by adding a previous batch of labeled data to training data and using a new batch of labeling data as test data to generate the series of NLU models with training data that increases constantly. If not all the labeling data is received, the method comprises repeating the step of building a series of NLU models until all labeling data is received. After all the training data is received, at least once, the method comprises building a third NLU model using all the labeling data, wherein the third NLU model is used in generating the spoken dialog service.

摘要翻译： 公开了一种生成在口头对话系统中使用的自然语言模型的方法。 该方法包括对标签指南中定义的每个呼叫类型使用样本话语和创建许多手工制作规则。 使用手工制作的规则和样品说话来生成和测试第一个NLU模型。 使用示例语句作为新的训练数据并使用手工制作规则构建了第二个NLU模型。 使用第一批标签数据对第二个NLU模型进行性能测试。 通过将前一批标签数据添加到训练数据并使用新批签名数据作为测试数据来生成一系列NLU模型，训练数据不断增加，构建了一系列NLU模型。 如果不是全部接收到标签数据，则该方法包括重复建立一系列NLU模型的步骤，直到接收到所有标记数据为止。 在接收到所有训练数据之后，至少一次，该方法包括使用所有标签数据构建第三NLU模型，其中第三NLU模型用于生成口语对话服务。

公开(公告)号：US07752036B2

公开(公告)日：2010-07-06

申请号：US12344815

申请日：2008-12-29

申请人： Richard Vandervoort Cox ,  Stephen Michael Marcus ,  Mazin G. Rahim ,  Nambirajan Seshadri ,  Robert Douglas Sharp

发明人： Richard Vandervoort Cox ,  Stephen Michael Marcus ,  Mazin G. Rahim ,  Nambirajan Seshadri ,  Robert Douglas Sharp

IPC分类号： G10L19/00

CPC分类号： G10L15/02 ,  G10L15/20

摘要： Recognizing a stream of speech received as speech vectors over a lossy communications link includes constructing for a speech recognizer a series of speech vectors from packets received over a lossy packetized transmission link, wherein some of the packets associated with each speech vector are lost or corrupted during transmission. Each constructed speech vector is multi-dimensional and includes associated features. After waiting for a predetermined time, speech vectors are generated and potentially corrupted features within the speech vector are indicated to the speech recognizer when present. Speech recognition is attempted at the speech recognizer on the speech vectors when corrupted features are present. This recognition may be based only on certain or valid features within each speech vector. Retransmission of a missing or corrupted packet is requested when corrupted values are indicated by the indicating step and when the attempted recognition step fails.

摘要翻译： 识别通过有损通信链路作为语音向量接收的语音流包括：通过有损分组化传输链路从分组接收的分组来构建语音识别器的一系列语音向量，其中与每个语音向量相关联的一些分组丢失或损坏 传输。 每个构造的语音向量是多维的并且包括相关联的特征。 在等待预定的时间之后，产生语音向量，并且在存在时将语音向量内潜在的损坏的特征指示给语音识别器。 当存在损坏的特征时，语音识别器在语音向量上尝试语音识别。 该识别可以仅基于每个语音向量内的某些或有效特征。 当指示步骤指示损坏的值以及尝试的识别步骤失败时，请求重新发送丢失或损坏的数据包。

公开(公告)号：US07624015B1

公开(公告)日：2009-11-24

申请号：US11276502

申请日：2006-03-02

申请人： Mazin G. Rahim ,  Giuseppe Riccardi ,  Jeremy Huntley Wright ,  Bruce Melvin Buntschuh ,  Allen Louis Gorin

发明人： Mazin G. Rahim ,  Giuseppe Riccardi ,  Jeremy Huntley Wright ,  Bruce Melvin Buntschuh ,  Allen Louis Gorin

IPC分类号： G10L15/14 ,  G10L15/18

CPC分类号： G10L15/142

摘要： A system and a method are provided. A speech recognition processor receives unconstrained input speech and outputs a string of words. The speech recognition processor is based on a numeric language that represents a subset of a vocabulary. The subset includes a set of words identified as being for interpreting and understanding number strings. A numeric understanding processor contains classes of rules for converting the string of words into a sequence of digits. The speech recognition processor utilizes an acoustic model database. A validation database stores a set of valid sequences of digits. A string validation processor outputs validity information based on a comparison of a sequence of digits output by the numeric understanding processor with valid sequences of digits in the validation database.

摘要翻译： 提供了一种系统和方法。 语音识别处理器接收无约束输入语音并输出一串字。 语音识别处理器基于代表词汇子集的数字语言。 该子集包括被识别为用于解释和理解数字串的一组单词。 数字理解处理器包含用于将字符串转换为数字序列的规则类型。 语音识别处理器使用声学模型数据库。 验证数据库存储一组有效的数字序列。 字符串验证处理器基于数字理解处理器输出的数字序列与验证数据库中的有效数字序列的比较来输出有效性信息。

公开(公告)号：US20090112585A1

公开(公告)日：2009-04-30

申请号：US12344815

申请日：2008-12-29

申请人： Richard Vandervoort Cox ,  Stephen Michael Marcus ,  Mazin G. Rahim ,  Nambirajan Seshadri ,  Robert Douglas Sharp

发明人： Richard Vandervoort Cox ,  Stephen Michael Marcus ,  Mazin G. Rahim ,  Nambirajan Seshadri ,  Robert Douglas Sharp

IPC分类号： G10L15/00

CPC分类号： G10L15/02 ,  G10L15/20

摘要： Recognizing a stream of speech received as speech vectors over a lossy communications link includes constructing for a speech recognizer a series of speech vectors from packets received over a lossy packetized transmission link, wherein some of the packets associated with each speech vector are lost or corrupted during transmission. Each constructed speech vector is multi-dimensional and includes associated features. After waiting for a predetermined time, speech vectors are generated and potentially corrupted features within the speech vector are indicated to the speech recognizer when present. Speech recognition is attempted at the speech recognizer on the speech vectors when corrupted features are present. This recognition may be based only on certain or valid features within each speech vector. Retransmission of a missing or corrupted packet is requested when corrupted values are indicated by the indicating step and when the attempted recognition step fails.

摘要翻译： 识别通过有损通信链路作为语音向量接收的语音流包括：通过有损分组化传输链路接收的分组来构建语音识别器的一系列语音向量，其中与每个语音向量相关联的一些分组丢失或损坏 传输。 每个构造的语音向量是多维的并且包括相关联的特征。 在等待预定的时间之后，产生语音向量，并且在存在时将语音向量内潜在的损坏的特征指示给语音识别器。 当存在损坏的特征时，语音识别器在语音向量上尝试语音识别。 该识别可以仅基于每个语音向量内的某些或有效特征。 当指示步骤指示损坏的值以及尝试的识别步骤失败时，请求重新发送丢失或损坏的数据包。

公开(公告)号：US07412383B1

公开(公告)日：2008-08-12

申请号：US10407965

申请日：2003-04-04

申请人： Tirso M. Alonso ,  Ilana Bromberg ,  Dilek Z. Hakkani-Tur ,  Barbara B. Hollister ,  Mazin G. Rahim ,  Giuseppe Riccardi ,  Lawrence Lyon Rose ,  Daniel Leon Stern ,  Gokhan Tur ,  James M. Wilson

发明人： Tirso M. Alonso ,  Ilana Bromberg ,  Dilek Z. Hakkani-Tur ,  Barbara B. Hollister ,  Mazin G. Rahim ,  Giuseppe Riccardi ,  Lawrence Lyon Rose ,  Daniel Leon Stern ,  Gokhan Tur ,  James M. Wilson

IPC分类号： G10L15/06 ,  G10L15/22

CPC分类号： G10L15/19 ,  G10L15/063 ,  G10L15/183 ,  H04M3/4936

摘要： Systems and methods for annotating speech data. The present invention reduces the time required to annotate speech data by selecting utterances for annotation that will be of greatest benefit. A selection module uses speech models, including speech recognition models and spoken language understanding models, to identify utterances that should be annotated based on criteria such as confidence scores generated by the models. These utterances are placed in an annotation list along with a type of annotation to be performed for the utterances and an order in which the annotation should proceed. The utterances in the annotation list can be annotated for speech recognition purposes, spoken language understanding purposes, labeling purposes, etc. The selection module can also select utterances for annotation based on previously annotated speech data and deficiencies in the various models.

摘要翻译： 用于注释语音数据的系统和方法。 本发明通过选择最有益的用于注释的话语来减少注释语音数据所需的时间。 选择模块使用包括语音识别模型和语言理解模型在内的语音模型来基于诸如由模型产生的置信度得分的标准来识别应当注释的话语。 这些话语被放置在注释列表中，以及要为语句执行的注释类型以及注释应该继续执行的顺序。 注释列表中的话语可以被注释用于语音识别目的，语言理解目的，标签目的等。选择模块还可以基于先前注释的语音数据和各种模型中的缺陷来选择用于注释的话语。

公开(公告)号：US07295981B1

公开(公告)日：2007-11-13

申请号：US10755014

申请日：2004-01-09

申请人： Narendra K. Gupta ,  Mazin G. Rahim ,  Gokhan Tur ,  Antony Van der Mude

发明人： Narendra K. Gupta ,  Mazin G. Rahim ,  Gokhan Tur ,  Antony Van der Mude

IPC分类号： G10L15/18

CPC分类号： G10L15/193 ,  G10L15/063 ,  G10L15/183

摘要： A method of generating a natural language model for use in a spoken dialog system is disclosed. The method comprises using sample utterances and creating a number of hand crafted rules for each call-type defined in a labeling guide. A first NLU model is generated and tested using the hand crafted rules and sample utterances. A second NLU model is built using the sample utterances as new training data and using the hand crafted rules. The second NLU model is tested for performance using a first batch of labeled data. A series of NLU models are built by adding a previous batch of labeled data to training data and using a new batch of labeling data as test data to generate the series of NLU models with training data that increases constantly. If not all the labeling data is received, the method comprises repeating the step of building a series of NLU models until all labeling data is received. After all the training data is received, at least once, the method comprises building a third NLU model using all the labeling data, wherein the third NLU model is used in generating the spoken dialog service.

摘要翻译： 公开了一种生成在口头对话系统中使用的自然语言模型的方法。 该方法包括对标签指南中定义的每个呼叫类型使用样本话语和创建许多手工制作规则。 使用手工制作的规则和样品说话来生成和测试第一个NLU模型。 使用示例语句作为新的训练数据并使用手工制作规则构建了第二个NLU模型。 使用第一批标签数据对第二个NLU模型进行性能测试。 通过将前一批标签数据添加到训练数据并使用新批签名数据作为测试数据来生成一系列NLU模型，训练数据不断增加，构建了一系列NLU模型。 如果不是全部接收到标签数据，则该方法包括重复建立一系列NLU模型的步骤，直到接收到所有标记数据为止。 在接收到所有训练数据之后，至少一次，该方法包括使用所有标签数据构建第三NLU模型，其中第三NLU模型用于生成口语对话服务。

公开(公告)号：US07292982B1

公开(公告)日：2007-11-06

申请号：US10447889

申请日：2003-05-29

申请人： Dilek Z. Hakkani-Tur ,  Mazin G. Rahim ,  Gokhan Tur

发明人： Dilek Z. Hakkani-Tur ,  Mazin G. Rahim ,  Gokhan Tur

IPC分类号： G06F17/21 ,  G06F17/27 ,  G10L15/08

CPC分类号： G10L15/1822

摘要： An active labeling process is provided that aims to minimize the number of utterances to be checked again by automatically selecting the ones that are likely to be erroneous or inconsistent with the previously labeled examples. In one embodiment, the errors and inconsistencies are identified based on the confidences obtained from a previously trained classifier model. In a second embodiment, the errors and inconsistencies are identified based on an unsupervised learning process. In both embodiments, the active labeling process is not dependent upon the particular classifier model.

摘要翻译： 提供了一种主动标注过程，其目的是通过自动选择可能是错误的或与先前标记的示例不一致的那些来最小化要再次检查的话语的数量。 在一个实施例中，基于从先前训练的分类器模型获得的信心来识别误差和不一致性。 在第二实施例中，基于无监督的学习过程来识别错误和不一致。 在两个实施方案中，活性标记过程不依赖于特定的分类器模型。

公开(公告)号：US07280965B1

公开(公告)日：2007-10-09

申请号：US10407565

申请日：2003-04-04

申请人： Lee Begeja ,  Richard Vandervoort Cox ,  Harris Drucker ,  David Crawford Gibbon ,  Allen Louis Gorin ,  Patrick Guy Haffner ,  Steven H. Lewis ,  Zhu Liu ,  Mazin G. Rahim ,  Bernard S. Renger ,  Behzad Shahraray

发明人： Lee Begeja ,  Richard Vandervoort Cox ,  Harris Drucker ,  David Crawford Gibbon ,  Allen Louis Gorin ,  Patrick Guy Haffner ,  Steven H. Lewis ,  Zhu Liu ,  Mazin G. Rahim ,  Bernard S. Renger ,  Behzad Shahraray

IPC分类号： G10L21/00

CPC分类号： G10L13/08 ,  G10L15/18

摘要： Systems and methods for monitoring labelers of speech data. To test or train labelers, a labeler is presented with utterances that have already been identified as belonging to a particular class or call type. The labeler is asked to assign a call type to the utterances. The performance of the labeler is measured by comparing the call types assigned by the labeler with the existing call types of the utterances. The performance of a labeler can also be monitored as the labeler labels speech data by occasionally having the labeler label an utterance that is already labeled and by storing the results.

摘要翻译： 用于监控语音数据标签器的系统和方法。 为了测试或训练标签商，标签器被呈现已经被识别为属于特定类别或呼叫类型的话语。 要求标签器为话语分配一个呼叫类型。 标签器的性能是通过将标签器分配的呼叫类型与话语的现有呼叫类型进行比较来测量的。 标签器的性能也可以被监视，因为标签器通过偶尔将标签标签标记为已经被标记的话语并且通过存储结果来标记语音数据。

公开(公告)号：US06775652B1

公开(公告)日：2004-08-10

申请号：US09107784

申请日：1998-06-30

申请人： Richard Vandervoort Cox ,  Stephen Michael Marcus ,  Mazin G. Rahim ,  Nambirajan Seshadri ,  Robert Douglas Sharp

发明人： Richard Vandervoort Cox ,  Stephen Michael Marcus ,  Mazin G. Rahim ,  Nambirajan Seshadri ,  Robert Douglas Sharp

IPC分类号： G10L1528

CPC分类号： G10L15/02 ,  G10L15/20

摘要： Recognizing a stream of speech received as speech vectors over a lossy communications link includes constructing for a speech recognizer a series of speech vectors from packets received over a lossy packetized transmission link, wherein some of the packets associated with each speech vector are lost or corrupted during transmission. Each constructed speech vector is multi-dimensional and includes associated features. Potentially corrupted features within the speech vector are indicated to the speech recognizer when present. Speech recognition is attempted at the speech recognizer on the speech vectors when corrupted features are present. This recognition may be based only on certain or valid features within each speech vector. Retransmission of a missing or corrupted packet is requested when corrupted values are indicated by the indicating step and when the attempted recognition step fails.

摘要翻译： 识别通过有损通信链路作为语音向量接收的语音流包括：通过有损分组化传输链路从分组接收的分组来构建语音识别器的一系列语音向量，其中与每个语音向量相关联的一些分组丢失或损坏 传输。 每个构造的语音向量是多维的并且包括相关联的特征。 语音向量中的潜在损坏的特征在存在时被指示给语音识别器。 当存在损坏的特征时，语音识别器在语音向量上尝试语音识别。 该识别可以仅基于每个语音向量内的某些或有效特征。 当指示步骤指示损坏的值以及尝试的识别步骤失败时，请求重新发送丢失或损坏的数据包。

公开(公告)号：US5806029A

公开(公告)日：1998-09-08

申请号：US528821

申请日：1995-09-15

申请人： Eric Rolfe Buhrke ,  Wu Chou ,  Mazin G. Rahim

发明人： Eric Rolfe Buhrke ,  Wu Chou ,  Mazin G. Rahim

IPC分类号： G10L15/02 ,  G10L15/14 ,  G10L15/20 ,  G10L5/00

CPC分类号： G10L15/144 ,  G10L15/02 ,  G10L21/0272

摘要： Hierarchical signal bias removal (HSBR) signal conditioning uses a codebook constructed from the set of recognition models and is updated as the recognition models are modified during recognition model training. As a result, HSBR signal conditioning and recognition model training are based on the same set of recognition model parameters, which provides significant reduction in recognition error rate for the speech recognition system.

摘要翻译： 分级信号偏移去除（HSBR）信号调理使用由该组识别模型构建的码本，并且在识别模型训练期间识别模型被修改时被更新。 因此，HSBR信号调理和识别模型训练基于相同的识别模型参数集，这显着降低了语音识别系统的识别误码率。