Patent US3909533A successfully refreshed from Google Patents.
US3909533A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Method and apparatus for the analysis and synthesis of speech signals
Assignee: OMNISEC AG TROCKENLOOSTRASSE 91 CH-8105 REGENSDORF SWITZERLAND A Co OF SWITZERLAND
•
Filing: 1974-10-08
•
Granted: 1975-09-30
•
Expires: 1994-10-08
Abstract
Abstract Synthesized speech is produced by a vocal tract model corresponding functionally to the human vocal tract and constructed of a linear digital filter. The parameter of the synthesis vocal tract model are determined by an analysis operation on an original speech signal using an identical vocal tract model, which may be the same model as used for synthesis. The analysis vocal tract model has its parameters adjusted according to a comparison between the original speech signal and the output signal of the analysis model so as to minimize the deviation between these two signals. Those parameters for which the deviation falls below a threshold value are used directly as parameters of the synthesis vocal tract model. The adjustments of the parameters are determined by a parameter computer working on the results of the aforementioned comparison and which itself may contain the vocal tract model.
Inventors
- 👤 Louis Sepp Willimann
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-10-24 20:35
CPC Classifications
G; G10; G10L; G10L25/00
Claims Analysis
Total Claims: 13
Claim 1
Independent Claim
336 words
In a method of analysing and synthesizing speech in which speech synthesis is effected by use of a synthesis vocal tract model which functionally corresponds to the human vocal tract and which is constructed essentially from a discrete linear filter, comprising: A. an analysis operation including the steps of 1. sampling an original speech signal and deriving therefrom 1a. a first group of signals representing parameters of said synthesis vocal tract model; 1b. a second group of signals representing the fundamental frequency reciprocal (hereinafter referred to as the ''''pitch period''''); and 1c. a third group of signals representing the voiced/voiceless character of each sample of the original speech signal; and B. A synthesis operation including the steps of 1. adjusting said synthesis vocal tract model by reference to said first group of signals, and 2a. during voiced samples of the original speech signal, exciting said synthesis vocal tract model by a train of pitch period spaced pulses, 2b. during voiceless samples of the original speech signal, exciting said synthesis vocal tract model by white noise, whereby a synthetic speech signal similar to said original speech signal is produced, The improvement wherein: C. said analysis operation (A) is performed by use of an analysis vocal tract model identical to said synthesis vocal tract model; D1. during voiced samples of the original speech signal said analysis vocal tract model is excited by a train of pitch period spaced pulses; D2. during voiceless samples of the original speech signal said analysis vocal tract model is excited by white noise; E. an output signal from the analysis vocal tract model is sampled and compared with the original speech signal; F. The parameters of the analysis vocal tract model are modified as a result of step (E) to minimize the deviation between the output signal and the original speech signal; and G. those parameters of said analysis vocal tract model for which the deviation falls below a predetermined threshold value are used as said first group of signals in step (A1a).
Claim 2
Dependent Claim
336 words
In a method of analysing and synthesizing speech in which speech synthesis is effected by use of a synthesis vocal tract model which functionally corresponds to the human vocal tract and which is constructed essentially from a discrete linear filter, comprising: A. an analysis operation including the steps of 1. sampling an original speech signal and deriving therefrom 1a. a first group of signals representing parameters of said synthesis vocal tract model; 1b. a second group of signals representing the fundamental frequency reciprocal (hereinafter referred to as the ''''pitch period''''); and 1c. a third group of signals representing the voiced/voiceless character of each sample of the original speech signal; and B. A synthesis operation including the steps of 1. adjusting said synthesis vocal tract model by reference to said first group of signals, and 2a. during voiced samples of the original speech signal, exciting said synthesis vocal tract model by a train of pitch period spaced pulses, 2b. during voiceless samples of the original speech signal, exciting said synthesis vocal tract model by white noise, whereby a synthetic speech signal similar to said original speech signal is produced, The improvement wherein: C. said analysis operation (A) is performed by use of an analysis vocal tract model identical to said synthesis vocal tract model; D1. during voiced samples of the original speech signal said analysis vocal tract model is excited by a train of pitch period spaced pulses; D2. during voiceless samples of the original speech signal said analysis vocal tract model is excited by white noise; E. an output signal from the analysis vocal tract model is sampled and compared with the original speech signal; F. The parameters of the analysis vocal tract model are modified as a result of step (E) to minimize the deviation between the output signal and the original speech signal; and G. those parameters of said analysis vocal tract model for which the deviation falls below a predetermined threshold value are used as said first group of signals in step (A1a).
Claim 3
Dependent Claim
53 words
A method as claimed in claim 1, in which step (F) includes determining the gradient of the error dimension representing the deviation with respect to the parameters of the analysis vocal tract model and modifying the parameters of the analysis vocal tract model in the opposite direction to the direction of the gradient.
Claim 4
Dependent Claim
48 words
A method as claimed in claim 2, in which following each determination of the erroR dimension representing the deviation between the original speech signal and the output signal of the analysis vocal tract model the parameters of the analysis vocal tract model are modified in a small step.
Claim 5
Dependent Claim
32 words
A method according to claim 3, in which the width of the step in the change of the parameters of the analysis vocal tract model is selected to have a fixed value.
Claim 6
Dependent Claim
34 words
A method as claimed in claim 1, in which in performing steps (D1) and (D2) said pulse train and said white noise, respectively, have a power which is substantially constant and substantially the same.
Claim 7
Dependent Claim
19 words
A method as claimed in claim 5, in which in performing step (D1) said pulse train comprises unit pulses.
Claim 8
Dependent Claim
159 words
Apparatus for performing speech analysis and synthesis comprising a synthesizer which includes a synthesis vocal tract model which functionally corresponds to the human vocal tract, and generator selectively operable to provide pulses or white noise; and an analyser including first means for determining parameters of said synthesis vocal tract model, second means for determining the pitch period of an original speech signal, and third means for determining the voiced/voiceless character of the original speech signal; wherein said first means comprises an analysis vocal tract model identical to the synthesis vocal tract model; a generator selectively operable to provide pulses or white noise identical to said generator of the synthesizer; a sample storage unit for storing samples of the original speech signal; a comparator for comparing an output signal of said analysis vocal tract model with the signal stored in the sample storage unit; and a parameter computer for minimizing the deviation between the two signals compared by said comparator.
Claim 9
Dependent Claim
159 words
Apparatus for performing speech analysis and synthesis comprising a synthesizer which includes a synthesis vocal tract model which functionally corresponds to the human vocal tract, and generator selectively operable to provide pulses or white noise; and an analyser including first means for determining parameters of said synthesis vocal tract model, second means for determining the pitch period of an original speech signal, and third means for determining the voiced/voiceless character of the original speech signal; wherein said first means comprises an analysis vocal tract model identical to the synthesis vocal tract model; a generator selectively operable to provide pulses or white noise identical to said generator of the synthesizer; a sample storage unit for storing samples of the original speech signal; a comparator for comparing an output signal of said analysis vocal tract model with the signal stored in the sample storage unit; and a parameter computer for minimizing the deviation between the two signals compared by said comparator.
Claim 10
Dependent Claim
26 words
Apparatus as claimed in claim 7, wherein said analysis vocal tract model and said synthesis vocal tract model are each comprised by a linear digital filter.
Claim 11
Dependent Claim
43 words
Apparatus as claimed in claim 8, wherein the parameter computer is adapted to be excited by the signal from said synthesizer generator and to provide an output signal corresponding to the gradient of the error dimension representing the deviation determined by said comparator.
Claim 12
Dependent Claim
136 words
Apparatus according to claim 9, wherein said parameter computer and said analyser vocal tract model are constituted by parts of a common unit which comprises: a primary model identical to said vocal tract model, a part of a model which is a dual model with respect to said primary model, and a number of additional part-models of the primary model corresponding to the number of components of the state vector of the primary model and of the dual part-model respectively; and wherein an input of said primary model and an input of said dual part-model are connected to the output of said synthesizer generator, and each of the additional primary part-models is connected by its input to each of those outputs of the dual part-model which yield the components of the state-vector of this dual part-model.
Claim 13
Dependent Claim
137 words
Apparatus as claimed in claim 9, wherein said parameter computer and said analyser vocal tract model are constituted by parts of a common unit which comprises: a primary model identical to said vocal tract model, a part of a model which is a first dual model with respect to said primary model, and a number of additional dual part-models corresponding to the number of components of the state vector of the primary model and of the dual part-model respectively; and wherein an input of said primary model and an input of said first dual part-model are connected to the output of said synthesizer generator and each of the other dual part-models is connected by its input to one of those outputs of the primary model which yield the components of the state vector of that primary part-model.
References Cited (Backward Citations)
US3631520A
Cited
US3624302A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 1978-09-29 | 1979-03-31 | 1979-09-30 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 1982-09-29 | 1983-03-31 | 1983-09-30 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 1986-09-29 | 1987-03-31 | 1987-09-30 | $7700.00 | Window Closed / Lapsed |