公开(公告)号：WO1994023426A1

公开(公告)日：1994-10-13

申请号：PCT/US1994002370

申请日：1994-03-07

Applicant: MOTOROLA INC.

Inventor： MOTOROLA INC. ,  GERSON, Ira, A. ,  JASIUK, Mark, A. ,  HARTMAN, Matthew, A.

IPC: G10L09/00

CPC classification number: G10L19/083 ,  G10L19/06 ,  G10L19/135 ,  G10L25/06 ,  G10L25/24 ,  G10L2019/0013

Abstract: A Vector-Sum Excited Linear Predictive Coding (VSELP) speech coder (200) provides improved quality and reduced complexity over a typical speech coder. VSELP uses a codebook (201) which has a predefined structure such that the computations required for the codebook search process can be significantly reduced. This VSELP speech coder uses single or multisegment vector quantizer of the reflection coefficients based on a Fixed-Point-Lattice-Technique (FLAT). Additionally, this speech coder uses a pre-quantizer to reduce the vector codebook search complexity and a high-resolution scalar quantizer to reduce the amount of memory needed to store the reflection coefficient vector codebooks. Resulting in a high quality speech coder with reduced computations and storage requirements.

Abstract translation: 矢量和激励线性预测编码（VSELP）语音编码器（200）提供了比典型语音编码器更好的质量和更低的复杂度。 VSEL​​P使用具有预定义结构的码本（201），使得可以显着减少码本搜索处理所需的计算。 该VSELP语音编码器使用基于定点晶格技术（FLAT）的反射系数的单个或多个矢量量化器。 此外，该语音编码器使用预量化器来减少矢量码本搜索复杂度和高分辨率标量量化器，以减少存储反射系数矢量码本所需的存储量。 产生了一个高质量的语音编码器，减少了计算和存储要求。

公开(公告)号：EP0764319A1

公开(公告)日：1997-03-26

申请号：EP96910297.0

申请日：1996-01-29

Applicant: MOTOROLA, INC.

Inventor： KUSHNER, William, M. ,  SRENGER, Edward ,  HARTMAN, Matthew, A.

IPC: G10L15

CPC classification number: G10L15/142

Abstract: In a statistical based speech recognition system, one of the key issues is the selection of the Hidden Markov Model that best matches a given sequence of feature observations. The problem is usually addressed by the calculation of the maximum likelihood, ML, state sequence by means of a Viterbi or other decoder. Noise or inadequate training can produce an ML sequence associated with a Hidden Markov Model other than the correct model. The method of the present invention provides improved robustness by combining the standard ML state sequence score (416) with an additional path core (418) derived from the dynamics of the ML score as a function of time. These two scores, when combined, form a hybrid metric (420) that, when used with the decoder, optimizes selection of the correct Hidden Markov Model (422).

公开(公告)号：WO1996030895A1

公开(公告)日：1996-10-03

申请号：PCT/US1996000968

申请日：1996-01-29

Applicant: MOTOROLA INC.

Inventor： MOTOROLA INC. ,  KUSHNER, William, M. ,  SRENGER, Edward ,  HARTMAN, Matthew, A.

IPC: G10L09/00

CPC classification number: G10L15/142

Abstract: In a statistical based speech recognition system, one of the key issues is the selection of the Hidden Markov Model that best matches a given sequence of feature observations. The problem is usually addressed by the calculation of the maximum likelihood, ML, state sequence by means of a Viterbi or other decoder. Noise or inadequate training can produce an ML sequence associated with a Hidden Markov Model other than the correct model. The method of the present invention provides improved robustness by combining the standard ML state sequence score (416) with an additional path core (418) derived from the dynamics of the ML score as a function of time. These two scores, when combined, form a hybrid metric (420) that, when used with the decoder, optimizes selection of the correct Hidden Markov Model (422).

Abstract translation: 在基于统计的语音识别系统中，关键问题之一是选择与给定的特征观测序列最匹配的隐马尔可夫模型。 通常通过维特比或其他解码器的最大似然度ML，状态序列的计算来解决该问题。 噪音或训练不足可产生与正确模型以外的隐马尔可夫模型相关的ML序列。 本发明的方法通过将标准ML状态序列分数（416）与作为时间的函数的ML得分的动力学导出的附加路径核心（418）组合来提供改进的鲁棒性。 当组合时，这两个分数形成混合度量（420），当与解码器一起使用时，优化选择正确的隐马尔可夫模型（422）。

公开(公告)号：WO1994019790A1

公开(公告)日：1994-09-01

申请号：PCT/US1994000724

申请日：1994-01-18

Applicant: MOTOROLA, INC.

Inventor： MOTOROLA, INC. ,  GERSON, Ira, A. ,  JASIUK, Mark, A. ,  HARTMAN, Matthew, A.

IPC: G10L05/00

CPC classification number: G10L19/12

Abstract: Analysis by synthesis calculates a difference by subtracting (130) synthesized speech from input speech. The synthesized speech is formed by exciting long and short term filters (124, 126) with excitation vectors from a codebook store (114) which is searched by codebook generation (120). A weighting filter (132) is applied to the difference signal and the weighted difference is used to calculate an energy measure (134) which is used to control the codebook search (140). The weighting filter is an Rth-order filter controlled with calculated coefficients. The method for calculating coefficients models the frequency response of L Pth-order filters by a single Rth-order filter, where the order R

Abstract translation: 通过合成分析，通过从输入语音中减去（130）合成语音来计算差值。 合成语音由具有来自码本生成（120）搜索的码本存储（114）的激励矢量的激励长期和短期滤波器（124,126）形成。 将加权滤波器（132）应用于差分信号，并且使用加权差来计算用于控制码本搜索的能量测量（134）（140）。 加权滤波器是用计算系数控制的R阶滤波器。 计算系数的方法通过单个R阶滤波器模拟L P阶滤波器的频率响应，其中阶数R