-
公开(公告)号：US3196212A
公开(公告)日：1965-07-20
申请号：US15775561
申请日：1961-12-07
Applicant: IBM
Inventor： HORWITZ LAWRENCE P , JOSE REINES , SHELTON GLENMORE L
Abstract: 986, 520. Indicating extremes of amplitude. INTERNATIONAL BUSINESS MACHINES CORPORATION. Dec. 4, 1962 [Dec. 7, 1961], No. 45737/62. Heading G1U. [Also in Division G4] In speech recognition apparatus an indication of local amplitude extreme in a number of frequency channels is provided by applying each channel to a transistor adapted to respond to a local maximum, the output of any transistor which responds serving also to inhibit response of at least adjacent transistors. The speech signal from microphone 7 is applied to band-pass filters 8a-8n the outputs of which pass to detectors 10a-10n giving D.C. amplitudes proportional to the square of the input amplitudes. These signals are therefore measures of the power in the corresponding frequency band. The curve of Fig. 2 represents the power in the different frequency bands, each maximum being accompanied by smaller peaks on each side, relating to harmonics of the maximum frequency. The circuit is designed to respond only to the main maxima and to ignore all or some of the adjacent peaks. The outputs from detectors 10a- 10n are each applied to comparator transistors 11a-11n which tend to conduct. When the transistor, say 11b of the channel having a maximum, conducts the potential at point 13b increases and exceeds the amplitude of the adjacent peaks. Through resistors 14a and 14b the potential of points 13a and 13c is also raised sufficiently to prevent the transistors 11a and 11c conducting. At more remote points 13d-13n the effect of the conduction of transistor 11b is more attenuated due to the successive resistors 14c-14n and more remote transistors are not affected so that other local maximum may be detected in the same way. Those transistors which conduct energize relays 17a-17n through second transistors 16a-16n The output, Fig. 4 indicates the local maxima. These are called the "formants" of the signal. A modification having two comparator stages for detecting more widely differing maxima such as the one at f6.
-
公开(公告)号：DE1180560B
公开(公告)日：1964-10-29
申请号：DEJ0020283
申请日：1961-07-24
Applicant: IBM
Inventor： HORWITZ LAWRENCE P , SHELTON GLENMORE L
Abstract: 990,531. Automatic character reading. INTERNATIONAL BUSINESS MACHINES CORPORATION. May 30, 1962 [June 19, 1961], No. 20754/62. Heading G4R. In a character reading apparatus the character is scanned to derive signals representing black and white areas of the character, some or all of these signals being autocorrelated to provide a multiplicity of second signals representing at least parts of different autocorrelation orders of the character pattern and means are provided for identifying the scanned character from the second signals. A first order autocorrelation function gives the number of pairs of black areas separated by a given distance in a given direction over all possible distances and directions. The function is derived by comparing pairs of points having the same positional relationship all over the character pattern and counting all pairs where both are black. The counts may be tabulated for each positional relationship. A second order autocorrelation function comparisons are made between groups of three points and counting all triples where all three are black. The process may be repeated for any number of points in a group and the general case is considered of autocorrelation functions of first, second, and so on up to the nth order. The character 1, Fig. 1, is scanned in a horizontal raster by a C.R.T. 5 and the reflected light, received in photo-cell 7, provides signals f(t) representing the character as shown in Fig. 5. Since the speed of scanning is uniform the character may be considered as being represented by a series of fortyfive pulses i.e. a function of time. The f(t) signals are combined in pairs, triples &c. in an N-tuple generator 9 and the outputs applied to identification circuit 11. The N-tuples are extracted as shown in Figs. 28a, 28b, 28c, the photo-cell output being sampled at 45 points in the scan, to obtain the series shown in Fig. 5, and applied to a shift register 125. At each step the signals in the register stages are gated, in gates 127, with the incoming signal. This compares each position in the signal with each other position and black-black coincidences appear as output pulses which are counted by being integrated in integrators 151. The output from the 2-element combination gates are applied to further gates 129 also connected to stages of the shift-register, thereby obtaining 3- element combinations and so on. The coincidences are counted as before. The integrator outputs are D.C. voltages each representing a term in the first, second, --nth order autocorrelation functions. The first and second order functions are shown in Fig. 9 for the character "3". The first order terms are indicated along the bottom edge and again on the diagonal. Only half the table is shown since the other side is a mirror-image. The values in the table indicate the number of coincidences of the original signal train with both of points t 1 and t 2 which vary from 0 to 22. Shaded areas are points which are on the fringe of the pattern area and can be ignored. The integrator outputs after amplification at 153 are applied via resistors to certain ones of character leads SSR1-SSR0. The connections are designed so that an ideal character gives a maximum output on the corresponding lead. The lead signals are normalised for area of character by weighted resistors 157 and applied to a transistor circuit which determines the most positive signal. In this circuit each lead is connected to an N-P-N transistor with a common emitter lead. Current flows only in the transistor connected to the highest signal. The conducting transistor operates a relay and lights a lamp. In the form of Fig. 3 the integrator outputs are applied to circuits which derive "entropy" functions E1, E2, E3 (Fig. 9) which represent the "order" or "disorder" of the pattern sensed. The functions are represented by D.C. voltages and are applied through suitably weighted resistors to the character leads as before. In another form certain combinations only of the character positions are compared. These combinations may represent certain shape elements, e.g. a horizontal line. The combinations are stored on flip-flops and the outputs gated to identify the character. In another arrangement successive shape elements operate successive flipflops in recognition chains, one for each character. The first chain to be fully operated identifies the character. Seven shape elements may be specified by the combinations selected e.g. those shown in Fig. 34a. In a last embodiment the shape element signals are sampled at six instants during the scanning starting with the occurrence of the first shape element. The expected occurrences of the shape elements for the ten characters 0-9 is shown in Fig. 34c. The same pattern derived from the scanned character is entered with seven shiftregisters in similar form and gates are provided which compare particular pairs of positions in the same and different columns as the pattern is entered. Coincidences are added in an integrator and the voltages derived are compared after normalisation to obtain the highest. This identifies the character as before. Specification 982,989 is referred to.
-
公开(公告)号：CA939544A
公开(公告)日：1974-01-08
申请号：CA94002
申请日：1970-09-24
Applicant: IBM
Inventor： PENNINGTON KEITH S , SHELTON GLENMORE L , WILL PETER M
IPC: H04N7/12
-
-
Search Results
3
records
(took 0.026 seconds)
