公开(公告)号：WO1998021711A1

公开(公告)日：1998-05-22

申请号：PCT/DE1997002429

申请日：1997-10-21

Applicant: ROBERT BOSCH GMBH ,  WIETZKE, Joachim ,  OPATZ, Michael

Inventor： ROBERT BOSCH GMBH

IPC: G10L05/06

CPC classification number: B60R16/0373 ,  G10L15/22

Abstract: Disclosed is a method for entering vocal orders, whereby each command produces a data output for the user, thereby acknowledging or rejecting the command sequence.

Abstract translation: 公开了一种用于经由建议的语言是由一个输出，其中，每个指令之后，以使指令序列可以确认或丢弃所述用户输入命令的方法。

公开(公告)号：WO1997017694A1

公开(公告)日：1997-05-15

申请号：PCT/EP1995004337

申请日：1995-11-04

Applicant: INTERNATIONAL BUSINESS MACHINES CORPORATION ,  KUNZMANN, Siegfried ,  MOHR, Karlheinz ,  BANDARA, Upali ,  LEWIS, Burn, L.

Inventor： INTERNATIONAL BUSINESS MACHINES CORPORATION

IPC: G10L05/06

CPC classification number: G10L15/197 ,  G10L15/183

Abstract: Disclosed are a method and an apparatus for adapting, particularly reducing, the size of a language model, which comprises word n-grams, in a speech recognition system. The invention provides a mechanism to discard those n-grams for which the acoustic part of the system requires less support from the language model to recognize correctly. The proposed method is suitable for identifying those trigrams in a language model for the purpose of discarding during the built-time of the system. Provided is also another automatic classification scheme for words which allows the compression of a language model, but under retention of accuracy. Moreover it allows an efficient usage of sparsely available text corpora because even singleton trigrams are used when they are helpful. No additional software tools are needed to be developed because the main tool, the fast match scoring, is a module readily available in the known recognizers themselves. Further improvement of the method is accomplished by classification of words according to the common text in which they occur as far as they distinguish from each other acoustically. The invention opens the possibility to make speech recognition available in low-cost personal computers (PC's), even in portable computers like Laptops.

Abstract translation: 公开了一种用于在语音识别系统中适应，特别是减小包括单词n-gram的语言模型的尺寸的方法和装置。 本发明提供了一种机制，用于丢弃系统声学部分所需的较少的语言模型识别正确识别的那些n克。 所提出的方法适用于识别语言模型中的三元组，以便在系统的内置时间内丢弃。 还提供了允许压缩语言模型但保持准确性的单词的另一种自动分类方案。 此外，它允许有效地使用稀疏可用的文本语料库，因为即使使用单例三元组也是有用的。 不需要额外的软件工具来开发，因为主要工具，快速匹配评分，是已知识别器本身容易获得的模块。 方法的进一步改进是通过根据它们发生的共同文本对词进行分类来实现的，只要它们在声学上彼此区分即可。 本发明开启了在低成本个人计算机（PC）中进行语音识别的可能性，即使在诸如笔记本电脑的便携式计算机中也是如此。

公开(公告)号：WO1996022514A3

公开(公告)日：1996-07-25

申请号：PCT/US1996000762

申请日：1996-01-19

Applicant: SRI INTERNATIONAL

Inventor： SRI INTERNATIONAL ,  DIGALAKIS, Vassilios ,  NEUMEYER, Leonardo ,  RTISCHEV, Dimitry

IPC: G10L05/06

公开(公告)号：WO1996013827A1

公开(公告)日：1996-05-09

申请号：PCT/GB1995002563

申请日：1995-11-01

Applicant: BRITISH TELECOMMUNICATIONS PUBLIC LIMITED COMPANY ,  RINGLAND, Simon, Patrick, Alexander

Inventor： BRITISH TELECOMMUNICATIONS PUBLIC LIMITED COMPANY

IPC: G10L05/06

CPC classification number: G10L15/063 ,  G10L2015/025

Abstract: A speech recogniser in which the recognition vocabulary is generated from a user's own speech by forming phonemic transcriptions of the user's utterances and using these transcriptions for future recognition purposes. The phonemic transcriptions are generated using a loosely constrained network, preferably one constrained only by noise. The resulting transcriptions therefore bear close resemblance to the user's input speech but require significantly reduced storage requirements compared to known speaker dependent word representations.

Abstract translation: 一种语音识别器，其中通过形成用户话语的音位转录和使用这些转录以用于将来的识别目的，从用户自己的语音产生识别词汇。 使用松散约束的网络产生音素转录，优选地仅由噪声约束的一个。 因此，所得到的转录与用户的输入语音密切相似，但与已知的与扬声器相关的词表示相比，需要显着降低存储要求。

公开(公告)号：WO1996010818A1

公开(公告)日：1996-04-11

申请号：PCT/US1995010552

申请日：1995-08-17

Applicant: MOTOROLA, INC.

Inventor： MOTOROLA, INC. ,  WANG, Shay-Ping, Thomas

IPC: G10L05/06

CPC classification number: G10L15/05 ,  G10L15/02 ,  G10L15/04 ,  G10L15/06 ,  G10L25/18 ,  G10L25/24 ,  G10L25/30

Abstract: Boundaries of spoken sound in continuous speech are identified by classifying delimitative sounds to provide improved performance in a speech recognition system (200). Delimitative sounds, those portions of continuous speech that occur between spoken sounds, are recognized by the same method used to recognize spoken sounds. Recognition of delimitative sounds is accomplished by training a learning machine (176) to act as a classifier which implements a discriminant function based on a polynomial expansion.

Abstract translation: 通过对分隔声音进行分类来识别连续语音中的声音边界，以在语音识别系统（200）中提供改进的性能。 声音中的声音，声音之间发生的连续语音部分被识别为用于识别语音的相同方法。 通过训练学习机（176）作为基于多项式展开实现判别函数的分类器来实现对分隔声音的识别。

公开(公告)号：WO1994023424A1

公开(公告)日：1994-10-13

申请号：PCT/GB1994000704

申请日：1994-03-31

Applicant: BRITISH TELECOMMUNICATIONS PUBLIC LIMITED COMPANY ,  SMYTH, Samuel, Gavin

Inventor： BRITISH TELECOMMUNICATIONS PUBLIC LIMITED COMPANY

IPC: G10L05/06

CPC classification number: G10L15/08 ,  G10L15/142

Abstract: A path link passing speech recognition system and method for recognising input connected speech, the recognition system having a plurality of vocabulary nodes (24) associated with word representation models, at least one of the vocabulary nodes (24) of the network being able to process more than one path link simultaneously, so allowing for more than one recognition result.

Abstract translation: 一种用于识别输入连接语音的路径链接传递语音识别系统和方法，所述识别系统具有与词表示模型相关联的多个词汇节点（24），网络的至少一个词汇节点（24）能够处理 同时多个路径链接，因此允许多个识别结果。

公开(公告)号：WO1994022132A1

公开(公告)日：1994-09-29

申请号：PCT/GB1994000629

申请日：1994-03-25

Applicant: BRITISH TELECOMMUNICATIONS PUBLIC LIMITED COMPANY ,  PAWLEWSKI, Mark ,  TANG, Joseph, Gordon

Inventor： BRITISH TELECOMMUNICATIONS PUBLIC LIMITED COMPANY

IPC: G10L05/06

CPC classification number: G10L17/02

Abstract: Apparatus for speaker recognition which comprises means (210, 220, 230) for generating, in response to a speech signal, a plurality of feature data comprising a series of coefficient sets, each set comprising a plurality of coefficients indicating the short term spectral amplitude in a plurality of frequency bands, and means (260) for comparing said feature data with predetermined speaker reference data, and for indicating recognition of a corresponding speaker in dependence upon said comparison; characterised in that said frequency bands are unevenly spaced along the frequency axis, and by means (250) for deriving a long term average spectral magnitude of at least one of said coefficients; and for normalising the or each of said at least one coefficient by said long term average.

Abstract translation: 用于说话者识别的装置，其包括用于响应于语音信号而产生包括一系列系数组的多个特征数据的装置（210,220,230），每个特征数据包括指示短期频谱幅度的多个系数 多个频带，以及用于将所述特征数据与预定的说话者参考数据进行比较并根据所述比较指示相应说话者的识别的装置（260） 其特征在于，所述频带沿着频率轴不均匀地间隔，并且通过用于导出至少一个所述系数的长期平均频谱幅度的装置（250） 以及用所述长期平均值来对所述至少一个系数的每一个进行标准化。

公开(公告)号：WO1994017519A1

公开(公告)日：1994-08-04

申请号：PCT/KR1994000007

申请日：1994-01-28

Applicant: KOREA TELECOMMUNICATION AUTHORITY ,  LEE, Chong, Rak

Inventor： KOREA TELECOMMUNICATION AUTHORITY

IPC: G10L05/06

CPC classification number: G10L13/07 ,  G10L15/04

Abstract: The present invention relates to a speech synthesis and recognition system that reduces the amount of memory capacity for storing standard speech information, and improves the synthesized speech quality and the rate of speech recognition. The speech synthesis and recognition system has a memory with the stored demiphoneme data bisected with respect to a center of phoneme, and produces a synthesis speech signal by decoding demiphoneme data stored in the memory and concatenating the decoded demiphoneme data while generating a character train data for word, phrase, clause corresponding to speech signal by comparing the demiphoneme data stored in the memory with the speech signal.

Abstract translation: 本发明涉及一种减少用于存储标准语音信息的存储容量的语音合成和识别系统，并且提高了合成语音质量和语音识别率。 语音合成和识别系统具有存储的相对于音素中心二维的存储的虹吸数据的存储器，并且通过解码存储在存储器中的demiphoneme数据并连接解码的虹吸数据来产生合成语音信号，同时产生用于 通过将存储在存储器中的虹吸数据与语音信号进行比较来对应于语音信号的单词，短语，子句。

公开(公告)号：WO1992006468A1

公开(公告)日：1992-04-16

申请号：PCT/GB1991001681

申请日：1991-09-30

Applicant: ENSIGMA LIMITED ,  THE SECRETARY OF STATE FOR DEFENCE ,  CAREY, Michael, John ,  PARRIS, Eluned Sarah ,  BRIDLE, John, Scott

Inventor： ENSIGMA LIMITED ,  THE SECRETARY OF STATE FOR DEFENCE

IPC: G10L05/06

CPC classification number: G10L17/00

Abstract: Speaker verification is important in such applications as financial transactions which are to be carried out automatically by telephone. False acceptances of a speaker cause serious problems but so do frequent false rejections in view of the annoyance caused. Some of the problems of speaker verification are reduced in the invention by forming Hidden Markov Models (HMMs) for each of a mumber of words using features of utterances of these words from a large number of speakers. These models are known as world models. In addition for every person whose speech is to be recognised, one HMM is formed for each of the words as uttered by that person. These models are known as personal models. In verification a person is prompted to repeat a string of isolated or connected words (15) and features from each of these words are extracted (16). Next the probabilities that these features could have been generated by the world models for these words and by the personal model of that person are calculated, respectively (17 and 18) and these probabilities are compared (19) for each word. A decision (23) on verification is based on a poll (22) of these comparisons.

公开(公告)号：WO1991018386A1

公开(公告)日：1991-11-28

申请号：PCT/US1991003362

申请日：1991-05-14

Applicant: VCS INDUSTRIES, INC. d.b.a VOICE CONTROL SYSTEMS

Inventor： VCS INDUSTRIES, INC. d.b.a VOICE CONTROL SYSTEMS ,  HUNT, Alan, K. ,  SCHALK, Thomas, B.

IPC: G10L05/06

CPC classification number: G10L17/24 ,  G07C9/00158 ,  G10L15/10 ,  G10L2015/0631 ,  H04M3/382 ,  H04M2201/40

Abstract: The present invention describes a method for recognizing alphanumeric strings spoken over a telephone network wherein individual character recognition need not be uniformly high in order to achieve high string recognition accuracy. Preferably, the method uses a processing system (fig. 2) having a digital processor (30), an interface (42) to the telephone network, and a database (45) for storing a predetermined set of reference alphanumeric strings. In operation, the system prompts the caller to speak each character of a string, beginning with a first character and ending with a last character. Each character is then recognized using a speaker-independent voice recognition algorithm. The method (fig. 6) calculates recognition distances between each spoken input character and the corresponding letter of digit in the same position within each reference alphanumeric string (108). After each character is spoken, captured and analyzed (105, 106), each reference string distance is incremented and the process is continued, accumulating distances for each reference string, until the last character is spoken. The reference string with the lowest cumulative distance (112) is then declared to be the recognized string.