US2803801A

US2803801A - Wave analyzing apparatus

Info

Publication number: US2803801A
Authority: US
Publication date: 1957-08-20
Anticipated expiration: 1974-08-20

Description

RSAAAc- Amm R. D. CUNNINGHAM WAVE ANALYZING APPARATUS 7 Sheets-Sheet 1 Filed April 1o. r1946 R. D. CUNNINGHAM' 2,803,801 wAvE ANALYZING APPARATUS" 7 Sheets-Sheet 2 Filed April 10, 1946 v ,wmbh

AM M NPV am@ N e fw *.N 7 M/A y@ P Aug. 20, 1957 R. D. CUNNINGHAM 2,803,801

WAVE ANALYZING APPARATUS Filed April 10, 1946 7 Sheets-Sheet 5 FIll- Aug. l20, 1957 R. D. UNNlNGHAM wAvE ANALyzING APPARATUS ug 20, 1957 R. D. CUNNINGHAM 2,803,801

WAVE ANALYZING APPARATUS i Allg 20, 1957 R. D. CUNNINGHAM WAVE ANALYZING APPARATUS ADA/54N A Ca/v/V//VGHAM R. DQ CUNNINGHAM WAVE ANALYZING APPARATUS Aug. 20, 1957 riledAprillo, 1946 7 Sheets-Sheet 7 www Unitd States Patent f" WAVE ANALYZING APPARATUS Rhean D. Cunningham, United States Army, Casper,

Wyo., assiguor to the United States of America as represented by the Secretary of War Application April 10, 1946, Serial No. 661,051

3 Claims. (Cl. 324-77) (Granted under Title 35, U. S. Code (1952), sec. 266) The inventiondescribed herein may be manufactured and used by or for the Government for governmental purposes, without the payment to me of any royalty thereon.

This invention is in electrical apparatus and more particularly provides a means and method for analyzing complex electrical signals, as speech signals. The circuit is specifically intended to detect and indicate the, location in the speech spectrum and the amplitude of the resonant bands present in speech vowels, the presence of consonants in speech signals, and the fundamental voiceifrequency.

It will be understood that speech is composed ofvowel and consonant sounds, the former predominating almost to the exclusion of the latter, that each of the vowel sounds contains two (occasionally three) relatively narrow resonant bands of variable location, spacing and amplitude, and that the consonant sounds are composed of energy which is distributed approximately evenly over the speech spectrum with some emphasis, however, on the upper half thereof.

The principal object of my invention is to provide novel means and method for analyzing complex electrical signals.

Another object is to provide novel means and method for analyzing a complex electrical signal to produce direct current voltages representing the frequencies of the resonant bands present in the signal.

A further object is to provide novel means and method for analyzing a complex electrical signal to produce direct current voltages representing the amplitudes of the resonant bands present in the signal.

An additional object is to provide a novel means and method for analyzing a complex electrical signal to indicate the presence of nonresonant components and their locations in the sound spectrum.

Other objects will be apparent from a reading of the following specification and claims.

In the drawings:

Figure 1 is a graphic representation of a vowel sound;

Figure 2 represents a vowel sound synthesized from three adjacent harmonically related signal frequencies of constant but different amplitudes; p

Figure 3 shows the wave of Figure 2 with pulse excitation of the three signal frequencies;

Figure 4 represents a typical consonant wave form;

Figures 5, 6, and 7 are schematic diagrams of portions of the circuit of my invention;

Figures 8 and 9 are block diagrams of other portions of my novel circuit; and

Figure 10 illustrates the invention as a whole by showing the relationships between Figures 5-9.

Referring to the drawings and particularly to Figure 1, a vowel sound may be seen to consist of a repetitive wave wherein appear, in addition to the fundamental frequency, two other frequencies, hereinafter referred to as formant frequencies, and so labelled in the drawing. The fundamental frequency shows as the repetition rate of the whole wave. One formant appears as a damped wave of several 2,803,801 Patented Aug. 20, 1957 times the fundamental frequency; the other formant, of much higher frequency, may be seen to be a damped wave superimposed upon the wave of the first formant.

The frequencies and amplitudes of the formants are determined by the harmonically related frequencies and amplitudes that make up the resonant band involved. When three adjacent harmonically related frequencies, each having a constant amplitude different from the others, are combined, a wave form such as is shown in Figure 2 is produced. If these frequencies are pulse excited (as in speech) at a fundamental frequency rate, so that there is a decay in amplitude toward zero after each excitation, a wave form such as is shown in Figure' 3 is produced. This is a formant wave. The frequency of each cycle of'this Wave varies from that of the others by a factor determined by the harmonically related frequencies involved and their respective amplitudes. The average cyclic frequency over an interval equal to the fundamental is the formant frequency. The amplitude of the formant is equal to the average amplitude of the cycles of the formant wave.

The consonants of speech produce wave forms similar to that shown in Figure 4. No definable formants appear since the frequencies comprising consonant sounds are not harmonically related and ordinarily cover large sections of the speech spectrum. Variations in amplitude are more or less erratic.

My present invention provides means based upon the foregoing considerations for analyzing speech, and hence providing indications of the frequencies and amplitudes of the formants present in vowel sounds, the presence and nature of consonants, and the fundamental frequency involved.

The circuit illustrated in the drawings and to be described includes three substantially like channels each of which furnishes two direct current outputs, one of which can be translated into an indication of the frequency of a resonant band present in the signal under examination and the other of which indicates the amplitude of the band and the fundamental frequency.

In the drawings, Figures 5 and 6 illustrate in detail one complete channel, namely, that intended to treat the highest resonant band, together with (in Figure 5) certain circuits which function intermediately of the top and middle channels for purposes later to be particularized, Figure 7 shows the initial stages of the second channel, together with circuitry functioning intermediately of the second and lowest channel, Figures 8 and 9 show the middle and low channels (as they cooperate with the input circuits found in Figures 5 and 7) in block form, and, as mentioned above, Figure 10 illustrates t-he manner in which Figures 5-9 should be related to obtain a complete showing of the apparatus. In the main, the description will be limited to that portion of the apparatus shown in Figures 5, 6, and 7, although brief comments will be made with respect to the other channels and the circuit as a whole. It will be understood that the three channels are essentially identical, differing only as hereinafter explained.

Referring now to Figures 5, 6, and 7, and particularly to Figure 5, it may be seen that the complex signal to be analyzed (herein assumed to be a speech signal) is introduced to the apparatus at 15 and amplified by pentode 16, the pentode having a load composed of a constant-current pentode 17 and an inductance 18. Two signals are derived, one from across the entire load (taken from the plate of amplifier 16), this having a nearly flat frequency response, and the other from across the inductance, this having a frequency response proportional t0 27rfL.

The speech signal obtained from across the inductance the inductance being essentially a slope filter) contains the amplitudes of the higher frequency formant waves as well as the amplitudes of the lower frequency formant waves, with the amplitudes of the higher frequency formant waves being accentuated with respect to the amplitudes of the lower frequency formant waves. This signal is introduced to a phase splitting circuit comprising a triode 20 having equal load resistances 21-22 in the plate and cathode circuits respectively. Both of the 180 phase difference signals are differentiated, the differentiated voltages being, of course, maximum for maximum rate of change and minimum fo'r minimum rate of change, the points at which the differentiated voltages drop to zero corresponding to the cycle peaks of the highest formant wave involved. The differentiation operation effectively removes the lower frequen- `cy formant waves from the two signals.

Each of the differentiated voltages drives the grid of one of a pair of triodes 23-24, having a common plate load resistance. As the differentiated voltages are 180 out of phase and both drop to zero at the same time, the grid of one of the triodes, as 23, is driven positive while the grid of the other, 24, is driven negative. Triode 23 then conducts producing a voltage drop across resistor 25. This decreases the plate voltage of tube 24, and this, together with the negative voltage on its grid, drives tube 24 to cut-off. When the differentiated voltages are zero, both triodes conduct equally at a normal rate, and the total plate current is at minimum; therefore, the voltage drop across resistor 25 is at its minimum. The combined action of triodes 23-24 is that of a full wave rectifier on the differentiated voltages derived from phase splitting circuit 20.

The rectified voltage produced across the load resistance 25 is negative with relation to the zero reference level of rectification. The points at which the rectified voltage decreases to zero correspond in time to the cycle peaks of the highest frequency formant wave of the input speech signal. The rectified voltage decreases to zero at a repetition rate that is twice that of the formant wave fre quency and is frequency modulated to correspond to the frequency modulation of the formant wave over the fundamental frequency interval.

The rectified voltage is used to overdrive an amplifier 30 normally biased to cutoff. The output of this amplifier consists of constant amplitude frequency modulated pulses of double the formant frequency with the pulse width being variable. The pulses are differentiated at 31-32 and the negative peaks removed by a clipper stage 33 so that positive frequency modulated pulses having constant amplitude and width are obtained. The wave forms can be seen in relation to the various circuit components in Figure 8 which, however, should-be understood to represent the second or middle channel of the analyzer. These pulses drive a cathode follower 34 which leads into a 400 cycle low pass filter 35. This filter averages the energy of the pulses, the filter output being a direct current Voltage proportional to the pulse repetition frequency and therefore, also proportional to the frequency of the top resonant band.

To determine the amplitude of the formant of the top resonant band, voltage obtained from the output of the double triode rectifier 23-24 is inverted at 40 and used to drive the grid of a triode forming a part of a voltage differencing circuit comprising tubes 41-42 coupled together by a common cathode resistor 43, the plate of tube 42 being connected directly to a source of B-lvoltage and the plate of tube 41 being provided with a resistive plate load 44. A second voltage obtained from the output of filter 35 is used to drive the grid of tube 42. The voltage developed across resistor 43 by tube 42 is proportional to the voltage on the grid of the tube. The action of this voltage with that of the rectified signal voltage on the grid of tube 41 effectively differences the two voltages, the difference voltage appearing across the plate resistor 44.

The differencing operation cancels the effect of slope filter 18 on the amplitudes of the different frequencies of the speech signal. The difference Voltage is inverted at 45 and used to drive a cathode follower 46 working into a 400 cycle low pass filter 47 just as above described, the output of the filter being a direct current voltage proportional to the amplitude of the formant wave of the top resonant band. Since the formant Wave varies in amplitude at the fundamental frequency of the voice, the filter output varies at the fundamental frequency rate, and, therefore, provides, in addition to an indication of the amplitude of the formant of the top resonant band, n indication of the fundamental voice frequency.

To obtain the amplitude and frequency of the formant of the next lower resonant band, the top resonant band must be eliminated from the speech signal leaving the formant waves of the lower resonant band or bands. Two signal voltages are obtained from across the load of the input amplifier 16 0f channel 1 above described. One signal voltage is obtained from across the entire load, this having the flat frequency response mentioned, while the other is taken from across the slope filter part 18 of the load, the frequency response of which is equal to 21rL.

The first of these two signals is fed through buffer amplifier 50 and the second through a buffer amplifier 51, the first comprising a variable mu pentode, the gain of which is controlled by a direct current biasing voltage obtained from the frequency indicating output derived from filter 35 in channel 1. Since this direct current biasing voltage varies with the frequency of the formant Wave of the top resonant band, the amplification thereof Will also be proportional to the frequency of the top resonant band formant. With the amplification of the fiat frequency response signal controlled so that the amplitude of the top resonant band formant Wave is equal to the amplitude of the same formant wave of the signal having a frequency response equal to 21rfL, the high frequency formant wave will be cancelled out when the two signals are differenced at 52. The resulting signal comprises the lower frequency formant wave or Waves with amplitudes varying inversely with 21rfL due to the difference in the frequency response of the two signals. The difference signal is amplified at 53 and passed through a slope filter 54 so that the frequency response again becomes nearly linear. The same analytical process used in channel 1 to obtain direct current voltages indicating the frequency and amplitude of the formant of the top resonant band is used in channel 2 to obtain the desired information relative to the formant of the next lower resonant band. This will be described further in connection with the block diagram of Figure 8.

If a third resonant band is present in the speech being examined, the input to channel 3 is amplified at 66 and two signals taken from the load 67-68 thereof, these being fed through amplifiers 69 and 70 into differencing circuit 60, the formant wave of the middle resonant band being thus differenced out at 60 just as the top formant Wave was differenced out at 52. The frequency and amplitude of the formant of the lower resonant band are determined in the manner already described for the top and intermediate bands except that the signal is passed through only one slope filter; therefore, the frequency response of the formant wave of the low resonant band remains linear with respect to the input signal throughout its analysis. This makes unnecessary a voltage differencing circuit like circuit 41-42 and further, since there are no lower frequency waves present, no differentiating network is required before rectification.

Referring to Figure 8, the wave under consideration minus the uppermost formant, indicated at 101, is amplified at 100, split into two signals opposite in phase at 102, both being differentiated at 103, and full-wave rectified at 104, the resulting signal being indicated at 105. The rectified voltage is used to over drive an amplifier 106 thereby to produce a sequence of pulses 107 of constant amplitude and variable width. The pulses are differentiated at 108 (producing a wave form like 108'), and the negative pulses removed in clipper 109 so that a series of positive pulses 109', representing in frequency the frequency of the middle formant, results. The pulses are introduced to filter 110 which effectively averages the same and produces a direct current voltage such as is indicated at 111. The output of rectifier 104 also is fed into an inverter 112 and compared at 113 with the output of filter 110, the difference voltage being further inverted at 114 and averaged at filter 115 to produce a second direct current voltage such as is illustrated at 116.

Considering Figure 9, the remainder of the signal (the two upper formants having been eliminated) is amplified and slope-filtered at 120 (see 121), further amplified at 122, split as above at 123 into two components opposite in phase and rectified at 124, the output of the rectifier appearing as at 125. The rectified voltage is utilized once more to over drive an amplifier 126 to produce a sequence of square positive pulses 127 which are differentiated at 128, thereupon appearing as a pulse sequence 129. The negative pulses are removed at 130 (see 131) and averaged at 132 to produce a direct current voltage indication (133) of frequency of the lowest resonant band. As in the other channels, the output of the rectifier also is fed directly to an inverter 134 and thereafter averaged at 135 to produce an amplitude indication 136.

If there are only two resonant bands present in the speech, there will be no formant wave present in channel 3, and the outputs of this channel will both indicate zero.

Briefiy summarized and related to the various circuit components, the operation of the analyzer is as follows: a speech signal normally containing two or more formant frequencies is treated in slope filter 18 to suppress the amplitudes of all except the upper band frequencies. The output of the filter is split into two signals opposite in phase, and differentiated, which operation effectively eliminates the lower frequency elements remaining. The two signals are rectified by means of triodes 23-24, the rectied voltage being used to overdrive an amplifier and thereupon to produce square pulses varying in width but constant` in amplitude, the frequency thereof being twice that of the formant under investigation and, therefore, proportional thereto. A further differentiation at 31-32 produces a sharp positive and a sharp negative peak for each square pulse, the negative peaks are eliminated by clipper stage 33, and the result is a succession of substantially identical peaked pulses the frequency of which is proportional to the formant. These pulses are averaged by means of filter 35 and converted to a direct voltage also proportional to the frequency of the formant. In the next channel, the uppermost formant is eliminated, and the remaining signal analyzed as before.

In the third channel, the second formant is eliminated, and any remaining signal is analyzed as described, except for obvious simplifications resulting from the fact that no more than one formant can be present.

In each case, a direct current indication of the amplitude of the signal portion being studied is obtained by passing the signal through a low-pass filter after compensating for the suppressive effect of the slope lter thereon.

The circuits above described will treat consonant sounds in the same manner as vowel sounds, but, since there are no well defined wave forms present in consonant sounds, the consonant frequencies being more or less random (though predominating in the upper part of the speech spectrum), the output of channel 1 will indicate a high frequency and the average energy level of the consonant at that frequency. The second and third channels will also indicate frequencies nearly as high as indicated by the outputs of the first channel. Since the differencing circuits are not capable of effectively differencing the erratic consonant frequencies, the result is that all three channels will indicate high frequencies of nearly the same value; it is this condition which is utilized to indicate the presence of a consonant.

The above description is in specific terms. It should be understood that the invention is not limited to the exact structure shown and described, and for the true scope of the invention reference should be had to the appended claims.

I claim:

1. In an apparatus for analyzing a complex signal and providing direct current voltages indicative of the frequency and amplitude of any formant waves therein, means for suitably amplifying the signal to be analyzed, slope filter means for accentuating the uppermost formant wave present with respect to any other waves, means for splitting the resulting signal into two similar components opposite in phase, means for simultaneously differentiating the said components, means for full-wave rectifying the differentiated voltages, means including an overdriven amplifier, a differentiating network, and a clipper circuit for converting the rectied voltage to a succession of frequency modulated pulses of constant amplitude and width, and means including a low-pass filter for averaging the energy of the pulses thereby to provide a direct current signal proportional to the frequency of the said uppermost formant wave.

2. The invention of claim 1 further characterized by means for differencing said rectified voltage and said last mentioned signal, and means for averaging the value of the differenced voltage, thereby to produce a direct current voltage proportional to the amplitude of the uppermost formant wave.

3. The invention of claim 1 further characterized byv means for eliminating said uppermost formant wave from said complex signal, and means for analyzing the remainder of said signal to provide direct current voltages indicative of the frequency and amplitude of any other formant waves.

References Cited in the file of this patent UNITED STATES PATENTS 1,976,481 Castner Oct. 9, 1934 1,994,232 Schuck Mar. 12, 1935 2,159,790 Freystedt et al May 23, 1939 2,183,248 Riesz Dec. 12, 1939 2,226,238 Doba Dec. 24, 1940 2,396,531 Reiskind et al Mar. 12, 1946 2,403,982 Koenig July 16, 1946 2,403,983 Koenig July 16, 1946 2,403,984 Koenig et a1. July 16, 1946