US5481644A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Neural network speech recognition apparatus recognizing the frequency of successively input identical speech data sequences
Assignee: Seiko Epson Corp
•
Filing: 1993-08-06
•
Granted: 1996-01-02
•
Expires: 2013-08-06
Abstract
Abstract The speech recognition apparatus recognizes a frequency of successively input identical speech data sequences. The speech recognition apparatus includes a speech recognition non-layered neural network unit. Speech data sequence is inputted as feature vectors from a feature extracting unit. The neural network performs speech recognition and determines whether the input speech data sequence matches at least one predetermined speech data sequence. The neural network generates a speech recognition signal when the input speech data sequence matches the at least one predetermined speech data sequence. A recognition signal detecting unit outputs a reset instruction signal each time the neural network generates the speech recognition signal. An internal state value setting unit resets the neural network unit to an initial state each time the recognition signal detecting unit outputs the reset instruction signal. Since the neural network unit is reset each time the speech recognition signal is outputted, accurate detection can be achieved even when speech data sequence to be recognized is inputted successively.
Inventors
- 👤 Mitsuhiro Inazumi
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-09 17:50
CPC Classifications
G; G10; G10L; G10L15/00; G10L15/08; G10L15/16
Claims Analysis
Total Claims: 57
Claim 1
Independent Claim
273 words
A speech recognition apparatus which recognizes a continuous input speech data sequence, the speech recognition apparatus recognizing the frequency of successively input identical input speech data sequences, comprising: feature extracting means for extracting a sequence of feature vectors from the continuous input speech data sequence by segmenting the continuous input speech data sequence into a sequence of frames and converting each frame of the sequence of frames of the continuous input speech data sequence into the sequence of feature vectors; speech recognition single-layered neural network means coupled to the feature extracting means for recognizing the sequence of feature vectors of the input speech data sequence input from the feature extracting means based on at least one learned predetermined input speech data sequence and for generating a speech recognition signal when the continuous input speech data sequence matches the at least one learned predetermined input speech data sequence; recognition signal detection means coupled to the speech recognition neural network means for detecting the speech recognition signal generated by said speech recognition neural network means and for generating a reset instruction signal; and internal state value setting means coupled to the speech recognition neural network means and the recognition signal detection means for setting an internal state value of the speech recognition neural network means by storing an initial value of the internal state value of said speech recognition neural network means and setting the internal state value of said speech recognition neural network means to the initial value based on the reset instruction signal generated by the recognition signal detection means, wherein the initial value is one of a plurality of internal state values.
Claim 2
Dependent Claim
273 words
A speech recognition apparatus which recognizes a continuous input speech data sequence, the speech recognition apparatus recognizing the frequency of successively input identical input speech data sequences, comprising: feature extracting means for extracting a sequence of feature vectors from the continuous input speech data sequence by segmenting the continuous input speech data sequence into a sequence of frames and converting each frame of the sequence of frames of the continuous input speech data sequence into the sequence of feature vectors; speech recognition single-layered neural network means coupled to the feature extracting means for recognizing the sequence of feature vectors of the input speech data sequence input from the feature extracting means based on at least one learned predetermined input speech data sequence and for generating a speech recognition signal when the continuous input speech data sequence matches the at least one learned predetermined input speech data sequence; recognition signal detection means coupled to the speech recognition neural network means for detecting the speech recognition signal generated by said speech recognition neural network means and for generating a reset instruction signal; and internal state value setting means coupled to the speech recognition neural network means and the recognition signal detection means for setting an internal state value of the speech recognition neural network means by storing an initial value of the internal state value of said speech recognition neural network means and setting the internal state value of said speech recognition neural network means to the initial value based on the reset instruction signal generated by the recognition signal detection means, wherein the initial value is one of a plurality of internal state values.
Claim 3
Dependent Claim
169 words
A speech recognition apparatus according to claim 1, wherein said speech recognition neural network means is comprised of a plurality of interconnected dynamic neurons, each i-th dynamic neuron having a neuron internal state value X i , where i=1 to n and n is a natural number said neuron internal state value X i of each i-th dynamic neuron changing with time to a value determined by a function X i =G(X i ,Z ij ), where i and j=1 to n, Z ij is input data to the i-th dynamic neuron from a j-th neuron, each of said dynamic neurons generating an output value based on a function F(X i ) where X i is the i-th neuron internal state value, and wherein said internal state value setting means includes a buffer memory, the internal state value setting means storing the internal state value of each of said dynamic neurons in the buffer memory as the neuron initial value when said speech recognition neural network means is stable.
Claim 4
Dependent Claim
38 words
A speech recognition apparatus according to claim 2, wherein said recognition signal detecting means generates the reset instruction signal when the speech recognition signal is continuously generated by said speech recognition neural network means for a predetermined time.
Claim 5
Dependent Claim
25 words
A speech recognition apparatus according to claim 3, wherein said function X i =G(X i Z ij ) is ##EQU18## where τ is a constant.
Claim 6
Dependent Claim
24 words
A speech recognition apparatus according to claim 4, wherein said function F(X i ) of each of said dynamic neurons is a sigmoid function.
Claim 7
Dependent Claim
32 words
A speech recognition apparatus according to claim 5, wherein said input data Z ij to each of said dynamic neurons includes the output value of the same neuron multiplied by a weight.
Claim 8
Dependent Claim
38 words
A speech recognition apparatus according to claim 6, wherein said input data Z ij to each said dynamic neurons includes the outputs of other neurons, each of the outputs of the other neurons multiplied by a corresponding weight.
Claim 9
Dependent Claim
25 words
A speech recognition apparatus according to claim 7, wherein said input data Z ij to each of said dynamic neurons includes an external input value.
Claim 10
Dependent Claim
105 words
A speech recognition apparatus according to claim 8, wherein said speech recognition neural network means includes a plurality of input neurons for inputting the sequence of said feature vectors of said continuous input speech data sequence input from the feature extracting means, a first output neuron for generating a positive signal output only when the input speech data sequence matches one of the at least one learned predetermined input speech data sequence, and a second output neuron for generating a negative signal output only when the input speech data sequence does not match one of the at least one learned predetermined input speech data sequence.
Claim 11
Dependent Claim
24 words
A speech recognition apparatus according to claim 4, wherein said function F(X i ) of each of said dynamic neurons is a threshold function.
Claim 12
Dependent Claim
61 words
A speech recognition apparatus according to claim 3, wherein said function X i =G(X i Z ij ) is ##EQU19## where W ij is a connection strength connecting an output Y j of the j-th neuron to an input of the i-th neuron, D i is an external input value, τ is a constant and θ i is a bias value.
Claim 13
Dependent Claim
31 words
A speech recognition apparatus according to claim 4, wherein said function X i =G(X i Z ij ) is ##EQU20## where S is a sigmoid function and τ is a constant.
Claim 14
Dependent Claim
66 words
A speech recognition apparatus according to claim 3, wherein said function X i =G(X i Z ij ) is ##EQU21## where S is a sigmoid function, W ij is a connection strength connecting an output Y j of the j-th neuron to an input of the i-th neuron, D i is an external input value, τ is a constant and θ i is a bias value.
Claim 15
Dependent Claim
58 words
A speech recognition apparatus according to claim 3, wherein said speech recognition neural network means includes at least one input neuron for inputting the sequence of feature vectors of said input speech data sequence from the feature extracting means, and at least one recognition result output neuron generating the recognition result of said speech recognition neural network means.
Claim 16
Dependent Claim
25 words
A speech recognition apparatus according to claim 2, wherein said function X i =G(X i Z ij ) is ##EQU22## where τ is a constant.
Claim 17
Dependent Claim
61 words
A speech recognition apparatus according to claim 2, wherein said function X i =G(X i Z j ) is ##EQU23## where W ij is a connection strength connecting an output Y j of the j-th neuron to an input of the i-th neuron, D i is an external input value, τ is a constant and θ i is a bias value.
Claim 18
Dependent Claim
31 words
A speech recognition apparatus according to claim 2, wherein said function X i =G(X i Z ij ) is ##EQU24## where S is a sigmoid function and τ is a constant.
Claim 19
Dependent Claim
66 words
A speech recognition apparatus according to claim 2, wherein said function X i =G(X i Z ij ) is ##EQU25## where S is a sigmoid function, W ij is a connection strength connecting an output Y j of the j-th neuron to an input of the i-th neuron, D i is an external input value, τ is a constant and θ i is a bias value.
Claim 20
Dependent Claim
24 words
A speech recognition apparatus according to claim 2, wherein said function F(X i ) of each of said dynamic neurons is a sigmoid function.
Claim 21
Dependent Claim
24 words
A speech recognition apparatus according to claim 2, wherein said function F(X i ) of each of said dynamic neurons is a threshold function.
Claim 22
Dependent Claim
32 words
A speech recognition apparatus according to claim 2, wherein said input data Z ij to each of said dynamic neurons includes the output value of the same neuron multiplied by a weight.
Claim 23
Dependent Claim
39 words
A speech recognition apparatus according to claim 2, wherein said input data Z ij to each of said dynamic neurons includes the outputs of other neurons, each of the outputs of the other neurons multiplied by a corresponding weight.
Claim 24
Dependent Claim
25 words
A speech recognition apparatus according to claim 2, wherein said input data Z ij to each of said dynamic neurons includes an external input value.
Claim 25
Dependent Claim
57 words
A speech recognition apparatus according to claim 2, wherein said speech recognition neural network means includes at least one input neuron for inputting the feature vector of said continuous input speech data sequence from the feature extracting means, and at least one recognition result output neuron generating the recognition result of said speech recognition neural network means.
Claim 26
Dependent Claim
104 words
A speech recognition apparatus according to claim 2, wherein said speech recognition neural network means includes a plurality of input neurons for inputting the sequence of said feature vectors of said continuous input speech data sequence input from the feature extracting means, a first output neuron for generating a positive signal output only when the input speech data matches one of the at least one learned predetermined input speech data sequence, and a second output neuron for generating a negative signal output only when the input speech data sequence does not match one of the at least one learned predetermined input speech data sequence.
Claim 27
Dependent Claim
38 words
A speech recognition apparatus according to claim 1, wherein said recognition signal detecting means generates the reset instruction signal when the speech recognition signal is continuously generated by said speech recognition neural network means for a predetermined time.
Claim 28
Dependent Claim
24 words
A speech recognition apparatus according to claim 1, wherein the plurality of internal state values are generated by the speech recognition neural network means.
Claim 29
Dependent Claim
16 words
A speech recognition apparatus according to claim 27, wherein said internal state values are variable values.
Claim 30
Dependent Claim
16 words
A speech recognition apparatus according to claim 1, wherein said initial value is a variable value.
Claim 31
Dependent Claim
272 words
A speech recognition apparatus which recognizes a continuous input speech data sequence, comprising: feature extracting means for extracting a sequence of feature vectors from the continuous input speech data sequence by segmenting input the continuous speech data sequence into a sequence of frames and converting each frame of the sequence of frame of the continuous input speech data sequence into the sequence of feature vectors; first speech recognition neural network means coupled to the feature extracting means for recognizing the sequence of feature vectors of the continuous input speech data sequence input from the feature extracting means based on at least one learned predetermined input speech data sequence and for generating a first speech recognition signal when the continuous input speech data sequence matches the at least one learned predetermined input speech data sequence; recognition signal detection means coupled to the speech recognition neural network means for generating an action instruction signal upon each detection of the first speech recognition signal generated by said first speech recognition neural network means; second speech recognition neural network means for recognizing the sequence of feature vectors of the continuous input speech data sequence input from the feature extracting means based on the at least one learned predetermined input speech data sequence and for generating a second speech recognition signal, when said action instruction signal is generated and the input speech data sequence matches the at least one learned predetermined input speech data sequence; and output generating means for selecting one of the first and second speech recognition signals generated by said first and second speech recognition neural network means as an output speech recognition signal.
Claim 32
Dependent Claim
272 words
A speech recognition apparatus which recognizes a continuous input speech data sequence, comprising: feature extracting means for extracting a sequence of feature vectors from the continuous input speech data sequence by segmenting input the continuous speech data sequence into a sequence of frames and converting each frame of the sequence of frame of the continuous input speech data sequence into the sequence of feature vectors; first speech recognition neural network means coupled to the feature extracting means for recognizing the sequence of feature vectors of the continuous input speech data sequence input from the feature extracting means based on at least one learned predetermined input speech data sequence and for generating a first speech recognition signal when the continuous input speech data sequence matches the at least one learned predetermined input speech data sequence; recognition signal detection means coupled to the speech recognition neural network means for generating an action instruction signal upon each detection of the first speech recognition signal generated by said first speech recognition neural network means; second speech recognition neural network means for recognizing the sequence of feature vectors of the continuous input speech data sequence input from the feature extracting means based on the at least one learned predetermined input speech data sequence and for generating a second speech recognition signal, when said action instruction signal is generated and the input speech data sequence matches the at least one learned predetermined input speech data sequence; and output generating means for selecting one of the first and second speech recognition signals generated by said first and second speech recognition neural network means as an output speech recognition signal.
Claim 33
Dependent Claim
30 words
A speech recognition apparatus according to claim 30, wherein said second speech recognition neural network means is reset each time the second recognition signal is generated for a predetermined time.
Claim 34
Dependent Claim
121 words
A speech recognition apparatus according to claim 31, wherein said second speech recognition neural network means is comprised of a plurality of interconnected dynamic neurons, each i-th dynamic neuron having a neuron internal state value X i where i=1 to n and n is a natural number, said neuron internal state value X i of each i-th dynamic neuron changing with time to a value determined by a function X i =G(X i ,Z ij ) where i and j=1 to n and Z ij is input data to each of said i-th dynamic neurons from the j-th dynamic neuron, said dynamic neuron converting said internal state value X i into an output value determined by a function F(X i ).
Claim 35
Dependent Claim
61 words
A speech recognition apparatus according to claim 32, wherein said function X i =G(X i Z ij ) is ##EQU26## where W ij is a connection strength connecting an output of the j-th neuron Y j to an input of the i-th neuron, D i is an external input value, τ is a constant and θ i is a bias value.
Claim 36
Dependent Claim
31 words
A speech recognition apparatus according to claim 32, wherein said function X i =G(X i Z ij ) is ##EQU27## where S is a sigmoid function and τ is a constant.
Claim 37
Dependent Claim
66 words
A speech recognition apparatus according to claim 32, wherein said function X i =G(X i Z ij ) is ##EQU28## where S is a sigmoid function, W ij is a connection strength connecting an output Y j of the j-th neuron to an input of the i-th neuron, D i is an external input value, τ is a constant and θ i is a bias value.
Claim 38
Dependent Claim
24 words
A speech recognition apparatus according to claim 32, wherein said function F(X i ) of each of said dynamic neurons is a sigmoid function.
Claim 39
Dependent Claim
24 words
A speech recognition apparatus according to claim 32, wherein said function F(X i ) of each of said dynamic neurons is a threshold function.
Claim 40
Dependent Claim
32 words
A speech recognition apparatus according to claim 32, wherein said input data Z ij to each of said dynamic neurons includes the output value of the same neuron multiplied by a weight.
Claim 41
Dependent Claim
38 words
A speech recognition apparatus according to claim 38, wherein said input data Z ij to each said dynamic neurons includes the outputs of other neurons, each of the outputs of the other neurons multiplied by a corresponding weight.
Claim 42
Dependent Claim
25 words
A speech recognition apparatus according to claim 39, wherein said input data Z ij to each of said dynamic neurons includes an external input value.
Claim 43
Dependent Claim
104 words
A speech recognition apparatus according to claim 40, wherein said speech recognition neural network means includes a plurality of input neurons for inputting said sequence of feature vectors of said continuous input speech data sequence input from the feature extracting means, a first output neuron for generating a positive signal output only when the input speech data sequence matches one of the at least one learned predetermined input speech data sequence, and a second output neuron for generating a negative signal output only when the input speech data sequence does not match one of the at least one learned predetermined input speech data sequence.
Claim 44
Dependent Claim
58 words
A speech recognition apparatus according to claim 32, wherein said speech recognition neural network means includes at least one input neuron for inputting the sequence of feature vectors of said input speech data sequence from the feature extracting means and at least one recognition result output neuron generating the recognition result of said speech recognition neural network means.
Claim 45
Dependent Claim
25 words
A speech recognition apparatus according to claim 32, wherein said function X i =G(X i Z ij ) is ##EQU29## where τ is a constant.
Claim 46
Dependent Claim
122 words
A speech recognition apparatus according to claim 30, wherein each of said first and second speech recognition neural network means is comprised of a plurality of interconnected dynamic neurons, each dynamic neuron having a neuron internal state value X i where i=0 and n is a material number, said neuron internal state value X i of each i-th dynamic neuron changing with time to a value determined by a function X i =G(X i ,Z ij ) where i and j=1 to n and Z ij is input data to each of said i-th dynamic neurons from a j-th dynamic neuron, said dynamic neuron converting said internal state value X i into an output value determined by a function F(X i ).
Claim 47
Dependent Claim
25 words
A speech recognition apparatus according to claim 44, wherein said function X i =G(X i Z ij ) is ##EQU30## where τ is a constant.
Claim 48
Dependent Claim
61 words
A speech recognition apparatus according to claim 44, wherein said function X i =G(X i Z j ) is ##EQU31## where W ij is a connection strength connecting an output of the j-th neuron Y j to an input of the i-th neuron, D i is an external input value, τ is a constant and θ i is a bias value.
Claim 49
Dependent Claim
31 words
A speech recognition apparatus according to claim 44, wherein said function X i =G(X i Z ij ) is ##EQU32## where S is a sigmoid function and τ is a constant.
Claim 50
Dependent Claim
66 words
A speech recognition apparatus according to claim 44, wherein said function X i =G(X i Z ij ) is ##EQU33## where S is a sigmoid function, W ij is a connection strength connecting an output Y j of the j-th neuron to an input of the i-th neuron, D i is an external input value, τ is a constant and θ i is a bias value.
Claim 51
Dependent Claim
24 words
A speech recognition apparatus according to claim 44, wherein said function F(X i ) of each of said dynamic neurons is a sigmoid function.
Claim 52
Dependent Claim
24 words
A speech recognition apparatus according to claim 44, wherein said function F(X i ) of each of said dynamic neurons is a threshold function.
Claim 53
Dependent Claim
32 words
A speech recognition apparatus according to claim 44, wherein said input data Z ij to each of said dynamic neurons includes the output value of the same neuron multiplied by a weight.
Claim 54
Dependent Claim
38 words
A speech recognition apparatus according to claim 44, wherein said input data Z ij to each of said dynamic neurons includes the outputs of other neurons, each the outputs of the other neurons multiplied by a corresponding weight.
Claim 55
Dependent Claim
25 words
A speech recognition apparatus according to claim 43, wherein said input data Z ij to each of said dynamic neurons includes an external input value.
Claim 56
Dependent Claim
59 words
A speech recognition apparatus according to claim 44, wherein said speech recognition neural network means includes at least one input neuron for inputting the sequence of feature vectors of said continuous input speech data sequence from the feature extracting means, and at least one recognition result output neuron generating the recognition result of said speech recognition neural network means.
Claim 57
Dependent Claim
104 words
A speech recognition apparatus according to claim 43, wherein said speech recognition neural network means includes a plurality of input neurons for inputting said sequence of feature vectors of said continuous input speech data sequence input from the feature extracting means, a first output neuron for generating a positive signal output only when the input speech data sequence matches one of the at least one learned predetermined input speech data sequence, and a second output neuron for generating a negative signal output only when the input speech data sequence does not match one of the at least one learned predetermined input speech data sequence.
References Cited (Backward Citations)
US4947482A
Cited
EP0318858A2
Cited
US5285522A
Cited
US5014219A
Cited
US5040215A
Cited
US5046019A
Cited
US5182794A
Cited
US5285523A
Cited
US5247584A
Cited
EP0510632A2
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 1999-01-01 | 1999-07-03 | 2000-01-02 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2003-01-01 | 2003-07-03 | 2004-01-02 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2007-01-01 | 2007-07-03 | 2008-01-02 | $7700.00 | Window Closed / Lapsed |