US5285522A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Neural networks for acoustical pattern recognition
Assignee: University of Pennsylvania
•
Filing: 1991-10-08
•
Granted: 1994-02-08
•
Expires: 2011-10-08
Abstract
Abstract A machine for neural computation of acoustical patterns for use in real-time speech recognition, comprising a plurality of analog electronic neurons connected for the analysis and recognition of acoustical patterns, including speech. Input to the neural net is provided from a set of bandpass filters which separate the input acoustical patterns into frequency ranges. The neural net itself is organized into two parts, the first for performing the real-time decomposition of the input patterns into their primitives of energy, space (frequency) and time relations, and the second for decoding the resulting set of primitives into known phonemes and diphones. During operation, the outputs of the individual bandpass filters are rectified and fed to sets of neurons in an opponent center-surround organization of synaptic connections ("on center" and "off center"). These units compute maxima and minima of energy at different frequencies. The next sets of neurons compute the temporal boundaries ("on" and "off"), while the following sets of neurons compute the movement of the energy maxima (formants) up or down the frequency axis. Then, in order to recognize speech sounds at the phoneme or diphone level, the set of primitives belonging to the phoneme is decoded such that only one neuron or a non-overlapping group of neurons fire when a particular sound pattern is present at the input. The output from these neurons is then fed to an Erasable Programmable Read Only Memory (EPROM) decoder and computer for displaying in real-time a phonetic representation of the speech input.
Inventors
- 👤 Paul H. Mueller
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-10-29 19:01
CPC Classifications
G; G06; G06N; G06N3/00; G06N3/02; G06N3/06; G06N3/063; G06N3/065; G10; G10L; G10L15/00; G10L15/08; G10L15/16; G06F; G06F2218/00
Claims Analysis
Total Claims: 29
Claim 1
Independent Claim
119 words
A method of recognizing a pattern in a received signal, comprising the steps of: dividing said received signal into a plurality of frequency components; determining primitives representing the respective differential relationships between energy, frequency and time characteristics of said plurality of frequency components using a first neural network comprising a plurality of interconnected electronic neurons; and decoding said primitives into predetermined patterns using a second neural network comprising a plurality of interconnected electronic neurons, whereby each respective neuron of said second neural network is connected to said first neural network so as to receive those primitives which make up said respective neuron's predetermined pattern and is activated when said primitives making up said respective neuron's predetermined pattern are received.
Claim 2
Independent Claim
119 words
A method of recognizing a pattern in a received signal, comprising the steps of: dividing said received signal into a plurality of frequency components; determining primitives representing the respective differential relationships between energy, frequency and time characteristics of said plurality of frequency components using a first neural network comprising a plurality of interconnected electronic neurons; and decoding said primitives into predetermined patterns using a second neural network comprising a plurality of interconnected electronic neurons, whereby each respective neuron of said second neural network is connected to said first neural network so as to receive those primitives which make up said respective neuron's predetermined pattern and is activated when said primitives making up said respective neuron's predetermined pattern are received.
Claim 3
Dependent Claim
22 words
The method of claim 1, wherein said received signal is an acoustical signal and said predetermined patterns represent predetermined phonemes and diphones.
Claim 4
Dependent Claim
64 words
The method of claim 2, wherein said primitive determining step comprises the steps of determining a second derivative of energy with respect to frequency for each frequency component (d 2 E/ds 2 ), determining a first derivative of energy with respect to time for each frequency component (dE/dt), and determining a first derivative of frequency with respect to time for each frequency component (ds/dt).
Claim 5
Dependent Claim
67 words
The method of claim 3, comprising the further step of organizing said plurality of interconnected electronic neurons of said first neural network into a first group of electronic neurons for calculating d 2 E/ds 2 , a second group of electronic neurons for calculating dE/dt, and a third group of electronic neurons for calculating ds/dt, each of said groups of electronic neurons being interconnected to each other.
Claim 6
Dependent Claim
37 words
The method of claim 1, comprising the further steps of: translating the predetermined patterns from activated electronic neurons of said second neural network into symbols for display; and displaying said symbols on a display in real time.
Claim 7
Dependent Claim
120 words
A pattern recognition system for recognizing a pattern in a received signal, comprising: means for dividing said received signal into a plurality of frequency components; a first neural network comprising a plurality of interconnected electronic neurons for determining primitives representing the respective differential relationships between energy, frequency and time characteristics of said plurality of frequency components; and a second neural network comprising a plurality of interconnected electronic neurons for decoding said primitives into predetermined patterns, whereby each respective neuron of said second neural network is connected to said first neural network so as to receive those primitives which make up said respective neuron's predetermined pattern and is activated when said primitives making up said respective neuron's predetermined pattern are received.
Claim 8
Dependent Claim
120 words
A pattern recognition system for recognizing a pattern in a received signal, comprising: means for dividing said received signal into a plurality of frequency components; a first neural network comprising a plurality of interconnected electronic neurons for determining primitives representing the respective differential relationships between energy, frequency and time characteristics of said plurality of frequency components; and a second neural network comprising a plurality of interconnected electronic neurons for decoding said primitives into predetermined patterns, whereby each respective neuron of said second neural network is connected to said first neural network so as to receive those primitives which make up said respective neuron's predetermined pattern and is activated when said primitives making up said respective neuron's predetermined pattern are received.
Claim 9
Dependent Claim
22 words
The system of claim 6, wherein said received signal is an acoustical signal and said predetermined patterns represent predetermined phonemes and diphones.
Claim 10
Dependent Claim
58 words
The system of claim 7, wherein said first neural network determines a second derivative of energy with respect to frequency for each frequency component (d 2 E/ds 2 ), a first derivative of energy with respect to time for each frequency component (dE/dt), and a first derivative of frequency with respect to time for each frequency component (ds/dt).
Claim 11
Dependent Claim
64 words
The system of claim 8, wherein said plurality of interconnected electronic neurons of said first neural network are organized into a first group of electronic neurons for calculating d 2 E/ds 2 , a second group of electronic neurons for calculating dE/dt, and a third group of electronic neurons for calculating ds/dt, each of said groups of electronic neurons being interconnected to each other.
Claim 12
Dependent Claim
35 words
The system of claim 6, further comprising means for translating the predetermined patterns from activated electronic neurons of said second neural network into symbols for display and means for displaying said symbols in real time.
Claim 13
Dependent Claim
28 words
The system of claim 10, wherein said translating means comprises an erasable programmable read only memory (EPROM) decoder coupled to an output side of said second neural network.
Claim 14
Dependent Claim
26 words
The system of claim 6, wherein said dividing means comprises a set of bandpass filters for dividing said received signal into said plurality of frequency components.
Claim 15
Dependent Claim
94 words
The system of claim 6, wherein each of said plurality of interconnected electronic neurons comprises: an input port; a positive output port; a negative output port; a plurality of operational amplifiers connected between said input and output ports; biasing means for biasing said plurality of operational amplifiers; and means connected to said positive and negative output ports for amplifying an output of said plurality of operational amplifiers such that at unity gain the input/output voltage relation of the electronic neuron is linear up to a maximum voltage and linear down to a minimum voltage.
Claim 16
Dependent Claim
41 words
The system of claim 13, wherein each of said plurality of interconnected electronic neurons further comprises indicating means coupled to at least one of said positive and negative output ports to indicate whether the corresponding electronic neuron is on or off.
Claim 17
Dependent Claim
27 words
The system of claim 13, wherein the fan-out capability of each of said plurality of interconnected electronic neurons at unity transfer gain is greater than five hundred.
Claim 18
Dependent Claim
24 words
The system of claim 13, wherein said minimum and said maximum voltages and the minimum outputs at said minimum and maximum voltages are adjustable.
Claim 19
Dependent Claim
137 words
An acoustical pattern recognition system for recognizing a pattern in a received acoustical signal, comprising: a set of bandpass filters for dividing said received acoustical signal into a plurality of frequency components; a first neural network comprising a plurality of interconnected electronic neurons for determining primitives representing the respective differential relationships between energy, frequency and time characteristics of said plurality of frequency components; and a second neural network comprising a plurality of interconnected electronic neurons for decoding said primitives into phonemes and diphones, whereby each respective neuron of said second neural network is connected to said first neural network so as to receive those primitives which make up a particular phoneme which is defined by the phoneme's set of primitives, each electronic neuron being activated when said electronic neuron's phoneme is present in said received acoustical signal.
Claim 20
Dependent Claim
137 words
An acoustical pattern recognition system for recognizing a pattern in a received acoustical signal, comprising: a set of bandpass filters for dividing said received acoustical signal into a plurality of frequency components; a first neural network comprising a plurality of interconnected electronic neurons for determining primitives representing the respective differential relationships between energy, frequency and time characteristics of said plurality of frequency components; and a second neural network comprising a plurality of interconnected electronic neurons for decoding said primitives into phonemes and diphones, whereby each respective neuron of said second neural network is connected to said first neural network so as to receive those primitives which make up a particular phoneme which is defined by the phoneme's set of primitives, each electronic neuron being activated when said electronic neuron's phoneme is present in said received acoustical signal.
Claim 21
Dependent Claim
58 words
The system of claim 17, wherein said first neural network determines a second derivative of energy with respect to frequency for each frequency component (d 2 E/ds 2 ), a first derivative of energy with respect to time for each frequency component (dE/dt), and a first derivative of frequency with respect to time for each frequency component (ds/dt).
Claim 22
Dependent Claim
64 words
The system of claim 18, wherein said plurality of interconnected electronic neurons of said first neural network are organized into a first group of electronic neurons for calculating d 2 E/ds 2 , a second group of electronic neurons for calculating dE/dt, and a third group of electronic neurons for calculating ds/dt, each of said groups of electronic neurons being interconnected to each other.
Claim 23
Dependent Claim
38 words
The system of claim 17, further comprising means for translating the phonemes and diphones from activated electronic neurons of said second neural network into phonetic symbols for display and means for displaying said phonetic symbols in real time.
Claim 24
Dependent Claim
28 words
The system of claim 20, wherein said translating means comprises an erasable programmable read only memory (EPROM) decoder coupled to an output side of said second neural network.
Claim 25
Dependent Claim
83 words
An electronic neuron comprising: an input port; a positive output port; a negative output port; a plurality of operational amplifiers connected between said input and output ports; biasing means for biasing said plurality of operational amplifiers; and means connected to said positive and negative output ports for amplifying an output of said plurality of operational amplifiers such that at unity gain the input/output voltage relation of the electronic neuron is linear up to a maximum voltage and linear down to a minimum voltage.
Claim 26
Dependent Claim
83 words
An electronic neuron comprising: an input port; a positive output port; a negative output port; a plurality of operational amplifiers connected between said input and output ports; biasing means for biasing said plurality of operational amplifiers; and means connected to said positive and negative output ports for amplifying an output of said plurality of operational amplifiers such that at unity gain the input/output voltage relation of the electronic neuron is linear up to a maximum voltage and linear down to a minimum voltage.
Claim 27
Dependent Claim
41 words
The neuron of claim 22, wherein each of said plurality of interconnected electronic neurons further comprises indicating means coupled to at least one of said positive and negative output ports to indicate whether the corresponding electronic neuron is on or off.
Claim 28
Dependent Claim
27 words
The neuron of claim 22, wherein the fan-out capability of each of said plurality of interconnected electronic neurons at unity transfer gain is greater than five hundred.
Claim 29
Dependent Claim
24 words
The neuron of claim 22, wherein said minimum and said maximum voltages and the minimum outputs at said minimum and maximum voltages are adjustable.
References Cited (Backward Citations)
US3950733A
Cited
US4450530A
Cited
US4774677A
Cited
US4660166A
Cited
US4760604A
Cited
US4773024A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 1997-02-07 | 1997-08-09 | 1998-02-08 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2001-02-07 | 2001-08-09 | 2002-02-08 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2005-02-07 | 2005-08-09 | 2006-02-08 | $7700.00 | Window Closed / Lapsed |