US 1948973 A
Description (OCR text may contain errors)
Feb. 27, 1934. J, c. STEINBERG WAVE TRANSMISSION WITH NAHROWED BAND Filed '.Jne 10 1932 from instant to instant.
Patented Feb. 27, 1934 PATENT @Fries WAVE TRANSMISSION WITH NARROWED BAND John C. Steinberg, Sparta, N. J., assigner to Bell Telephone Laboratories, Incorporated, New York, N. Y., a corporation of New York Application June 10, 1932. Serial No. 616,435
s claims. (ci. 17a-44) The present invention relates to a method and system for transmitting intelligence in" which the total frequency range required is less than that embraced in the intelligence-conveying waves as produced.
The invention will be described with particular reference to speech waves, although the invention is equally applicable to other kinds of waves such as music as well as waves used to produce visual elects or the like.
The invention depends upon the existence of` certain statistical characteristics in the waves, which enable their subdivision into components of diierent amplitude and frequency level, varying The components occurring at any one small' instant of that time do not fill-the entire frequency range occupied by the waves occurring over a considerable time. Advantage of this fact is taken by shifting the frequency level of certain components where they exist substantially alone at any instant so that they occupy the same frequency level as (or at least overlap in frequency) the components of other frequencies existing substantially alone at some other instant of time.
In the case of speech, it has been determined that the lower portion of the total frequency range is predominantly occupied by the vowel and semi-vowel sounds. The upper portion is occupied principally by the stop and fricative consonant sounds. Moreover, both classes of sounds do not exist in speech in the same instant of time but they occur in rapid succession. This. fact makes it possible to divide the one class of sounds from the other on a time basis. Also the amplitudes of the vowel and semi-vowel sounds are larger than those of the stop and fricative consonants. The two classes can, therefore, be distinguished from each other in time, in frequency range and in amplitude or energy.
In practicing the invention, lters may be used to separate the low frequency-large energy components from the high frequency-low energy components. A device responding differently to these two classes of components, switches one or other of the lters into circuit. One filter passes the components directly to the line. The other leads to a frequency-shifting circuit `such as a modulator which changes the frequency level of these components so that they occupy the same range as those passed by the other lter, or at least overlap that range. Thus both classes of sounds are transmitted in a reduced band width.
At the receiver a second device differently responsive to the two classes of wave components controls connection of the line to a pair of circuits one of which leads directly to the receiver and the other to a frequency shifting circuit which eifects a change opposite to that produced in 60 these components at the transmitter.
A system of this general character is disclosed in my Patent 1,836,824, granted December 15, 1931. The present invention comprises an alternative method and means of distinguishing vowel o5 and semi-vowel sounds from consonant sounds. The invention will be illustrated and described as embodied in a speech transmission system which is identical with that disclosed in my patent supra except for the means employed to distinguish between the two classes of speech components. The invention is, of course, capable of other embodiments and may be used, for ex ample, merely to effect separation of the vowel and semi-vowel sounds from the consonant sounds without regard to the use for which such separation is made.
In accordance with the present-invention. that characteristic of speech is made use of which is exhibited in the fact that vowel and semi-vowel sounds contain large components which are harmonies of a fundamental frequency whereas stop and fricative consonant sounds either are lacking in these harmonics or contain them only in a much smaller degree.
The wave forms of the vowels and semi-vowels show that they are characterized by certain transient and resonant variations that are impressed on a direct stream of air that has been modulated with the fundamental or vocal cord tone. This no fundamental in the case of a. mans voice is of the order of 125 cycles a second while in a woman's voice it is around 250 cycles. The prominent harmonies lie in the frequency range betweenv 300 and 2000 cycles.
The wave forms of the stop and fricative consonants show that they also are characterized by transient and resonant variations which, however, are impressed on a direct stream of alr which appears to be modulated with variations 100 that arise from the frictional effects of the passage of air through the vocal cavities. In the case of the voiced stop and fricative consonants, the fundamental or vocal cord tone is also present to a small extent, but there is a noticeable ab- 105 sence of overtones of the fundamental in the frequency range from 300 to 2000 cycles.
The semi-vowel sounds which contain prominent harmonics in this range and thus resemble the vowels are l. 1n, n. no. r, w. and y. vThe unno land will be of insufiicient voiced stop and fricative consonants are p, k, t, ch, (chin) and f, s, sh, th. (thin) respectively. The` corresponding voiced consonants which, as stated, are weak in harmonics of the fundamental in the region 300 to 2000 cycles, are b, g, d, gi, and v, z, zh (azure), th (then). These latter are thus classed with the. corresponding unvoiced consonants as regards lack of prominent harmonics in the range mentioned.
In accordance with the present invention the two classes of sounds (vowels and semi-vowels as against stop and fricative consonants) are separated by detecting or rectifying speech waves in the region from about =300 cycles up to the order of 1700 to 2000 cycles and passing the result- `ant detectedcomponents through a filter with a pass range from about 100 to 300 cycles. The sounds that have prominent over-tones of a fundamental of either 125 cycles or .250 cycles will upon vdetection yield a considerable amount of such fundamental, which will pass through the lter and may be used to operate suitable switching or selecting apparatus. Sounds that are weak or lacking in these harmonics will produce little or no component in the 100l to 300 cycle range strength to operate the switching device.
The invention will now he described as embodied in a two-way telephone system shown schematically in the single iigure of the accompanying drawing, which as stated is identical with 3 of my patent supra except for the switch control circuits.
The drawing shows a two-way terminal circuit for interconnecting the subscribers line L1 at the left with the main telephone line L2 shown at the right.
It is assumed in this illustrative example that the maximum frequency which it is practicable to transmit-over the line La is about 2250 cycles. Of course, speech could be transmitted over this line employing the range from about 250 cycles to 2250 cycles and commercial quality for many purposes would be obtained. However, in case it is desired to obtain a higher quality of transmission than would be possible with the range 250 to 2250 cycles by direct transmission, the
method of the present invention permitting utilization of the range 250 to 3250 cycles per second will be described showing how an increase in quality i'or this available total line range can be obtained.
The subscribers line L1 and the telephone line La are each shown as provided with the usual hybrid coils H1 and H2 respectively, and the usual line balancing net works L1N and LzN respectively.
For transmitting from the line L1 to the line La, there is provided a path generally indicated by T and for transmitting in the opposite direction between Ia and L1 there is provided a second path generally indicated by R.
In the transmitting path T beginning at the hybrid coil H1 there is rst a volume control circuit for reducing to the same level speech waves received over different length subscribers lines or from individuals having different voice strength. This volume control may be of any suitable type, either automatic or manual, but preferably automatic, examples of this type being given in patents of R. C. Mathes No. 1,810,025 granted June 16, 1931, and D. Mitchell No. 1,853,- 070 granted April 12, 1932. 'Ihis volume control when used in the present system will have va relatively large time lag so that its adjustment will not be changed for instantaneous variations in speech level but so that it will regulate only the average speech level over a considerable period of time.
On the output side of the volume control 10 the circuit branches into the two portions 11 and 12.
Branch 11 leads through a lter 35, detector 36, lter 37 and rectifier 38 to the switching relay 14. Filter 35 has a pass range from about 300 to 1700 cycles, suilicient to embrace the principal harmonics of the fundamental or vocal cord frequency. Strong harmonics in this range are acted upon in detector 36 to yield an output component of fundamental frequency which is selectively passed through filter 37 and upon being rectified at 38, causes operation of relay 14. Suitable amplification (not shown) may be used in branch 1l. Thus vowel and-semi-vowel sounds will cause actuation of relay 14 while consonant sounds will fail to operate the relay.
Circuit 12 leads directly to filter 15, through the normally closed back contact of relay 14. This filter 15 is in the consonant branch, which branch is closed unless relay 14 is actuated by vowel sounds. Filter 15 is designed to pass currents of frequencies between 1250 cycles per second and 3250 cycles per second but to suppress currents of both lower and higher frequencies. This and the other filters throughout the system may be designed inaccordance with the principles laid down in the U. S. patent to G. A. Campbell 1,227,113, patented May 22, 1917.
The output side of lter 15 is connected to modulator 16 which is preferably of the type disclosed in U. S. patent to Carson 1,343,306, dated June 15, 1920. This modulator is supplied with carrier waves from source 17, these waves having a frequency of 20,000 cycles per second. This source is preferably a vacuum tube oscillator such as is commonly used in carrier wave systems. As explained in the Carson patent, this type of'modulator prevents the unmodulated carrier components from the source I7'from passing into the outgoing circuit so that only the two side bands resulting from the modulating action of the circuit appear in the final output.
'I'he output of modulator 16 lis connected through a second illter 18 which is designed to pas only the range from 21,250 to 23,250 cycles per second, this being the upper side band resulting from the modulation in circuit 16. This side band is then demodulated in circuit 19 by means o f a sustained wave of 21,000 cycle frequency generated at source 20. 'Ihis demodulator 19 may be identical with the modulator circuit 16 and is of the type disclosed in the Carson patent referred to. As the result of the demodulation in circuit 19, a lower side band extending from 250 cycles to 2,250 cycles per second is produced and this passes through the fllter 21, the other components of modulation being outside the transmission range of the filter.' and therefore suppressed. y
As a result of the steps performed by the apparatus 15 to 21 inclusive,it will be seen that the consonant frequencies in the range 1,250 to 3,250 cycles per second have been stepped down in frequency to occupy the range 250 to 2,250 cycles per second.
When relay\14 attracts itsarmature, in response to vowel components, circuit 12 is connected to the input of lter 22 which passes the range 250 to 2,250 cycles per second directly to the outgoing circuit 23. -After suitable amplification, the currents in circuit 23 are transmitted o income by way of vhybrid coil Hz into the outgoing line Le.
'The apparatus in the receiving side R is closely analogous to that in 'the transmitting branch 'I' and the description will follow readily from that which has been given for branch T.
In the'receiving branch, the elements 45, 00, 47 and 48 are analogous to elements 35. 30, 37 and 38, already described, and function in similar modulators 16 and 3l, and similarly, one source would be used in place of the two sources 20 and 29 shown on `the drawing. Filter 3e may be identical with iilter 22.
The operation of the circuit is as follows. Speech waves coming in over the subscribers line L1 pass through the volume'control 10 and are reduced under varying conditions of service to the same volume level. These waves pass in part into the branch 12 and in part into the'branch 11. 1f the waves in branch 11 contain insufdcient harmonic component to yield a requisite level of fundamental, relay 14 remains unactu= yated and branch 12 remains connected to the lter 15. .As explained above, the relay le is adjusted so that it attracts its armature in response to vowel and semi-vowel sounds but does not attract its armature in response to consonant wavesin the circuit 11. Consequently the consonant components in the subscribers speech occupying the range from 1,250 cycles to 3,250 cycles per second pass freely through the iilter 15 and into the modulator 16 where they modulate the sustained waves of 20,000 cycles from the source 17. The upper side band resulting from this modulation is passed through the illter 10 to .the demodulator 19 where it is combined with the sustained wave of 21,000 cycle frequency from the source 20. The lower side band of this demodulating process lies in the range from 250 to 2,500 cycles per second and therefore passes freely through the filter 21- and through the outgoing amplifier into the hybrid coil H2 and the outgoing line La. As a result of this process the consonant sounds which normally occupy the range from 1,250 to 3,250 cycles per second pass into the outgoing linel Lc in the reduced frequency range extending from250 -to 2,250 cycles persecond.
Vowel sound components and semi-vowel components present in the subscrlbers speech passing into the circuit 11v cause actuation of relay 14 and shift the circuit 12 from the input of nl ter 15 to the input of iilter 22 so that these vowel.
sounds and semi-vowel sounds which occupy the range from 250 to 2,250 cycles per second pass directly into the outgoingv branch 23 and so toune La.
Since normally e'ach word oi speech will comprise partly vowel or semi-vowel sounds and consonant sounds, the' relay 14 will be operating continually during sustained speech and for this reason its operation should bemade as fast as practicable in order not to seriously clip any portions of the words or syllables being spoken.
of the From the foregoing description it Will be evident that the vowel and the consonant sounds transmitted to the line Lz both occupy the same frequency range and it is necessary therefore to separate the vowel sounds from the consonant sounds in order to receive the speech in intelligible manner.
The station at the opposite end of the line L2 may be an exact duplicate of the station shown on the drawing, so that the mannerin which the speech waves are received at the distant terminal will be evident from considering that speech waves are coming in on the line L2 to the station shownin the drawing after having been treated at the distant station in exactly the same manner as has been described above.
As already stated, consonant over L2 from the distant station contain insuiiicient harmonics of the fundamental voice component to cause operation-of relay 26. lThese sounds therefore pass directly into the lter 2'! and from there .into the modulator 20 where they are stepped up in frequency by combining with the wave source 20. The upper side band resulting from this modulation is passed through the illter 30 and is demodulated with waves from the source 32 of 20,000 cycle frequency producing a side band. extending from 1,250 to 3,250 cycles per second. This side band is transmitted to the n lter 33 and into the'line L1. As a result oi the modulating actionjust described, it is the consonant sounds which were transmitted at reduced frequency level over the line La are elesounds' arriving oi 21,000 cycle frequency from seen that are transmitted into the subscribers lne Li.
Vowel and semi-vowel lsound components received over the line L2 actuate the relay 26 and connect the incoming circuit to the lter 34 instead or to the nlter 27. rli'hese vowel and semivowel soundcornponents occupying the range 250 to 2,250 cycles per second pass directly into the outgoing subscriber's line L1.
It will be seen that a savingl in the total frequency range of 1,000 cycles per second has been eilected since the components in the range 250 to 3,250 cycles per second have been transmitted in vthe total range of 25,0 to 2,250 cycles per second the vowel and semi-vowels on the one hand and consonant sounds on the other, and the frequency level of the consonant components has been stepped downward toincrease the overlap with respect to the vowel sounds. It is within the invention, or course, to shift the frequency level vowel sound components to increase the normal overlap with respect to the consonant components.
The invention is not to be construed as limited have been given by way ot example, nor to the specific circuits that have been illustrated. For example, if the line La were capable oftransmitting the range 250 to 2,750 cycles per second, the division .might be as follows;
Vowel and semi-vowel sounds-250 to 2,050 cycles per second f Consonant second.
Various other frequency limits and frequency to particular frequency values or ranges that uas-1.2m to 3,750 cycles per '145 divisions will occur to any one desiring to practice the invention in connection with a particular system or situation;
It will be noted that the present invention depends not merely upon amplitude of the speech waves but upon the presence or absence of certain frequency relations existing in some sounds and not in others. The existence or non-existence of these frequency relations is independent of absolute volume. The method of the present invention offers an advantage in distinguishing from noise currents which may lie in the vowel range and have large amplitude but do not exhibit the characteristic of having prominent components harmonically related to the speech fundamental. The method of the invention also, and in the same way distinguishes from strongly inflected or emphasized consonants, such as sh may have amplitudes comparable with vowel sounds but lack the characteristic harmonically related components of vowels. Also the volume may vary to a greater extent than where relative amplitudes alone are relied upon, without interfering with the operation of the method according to this invention. l
Thus, in accordance with the present invention, the unvoiced consonant sounds may be amplified to any desired level before transmission so as to enable them to over-ride noise on the line. This amplification does not interfere with separation of ,the vowel and consonant sounds at the resonant branch at 49 for producing a controllable amount of amplification before the line is reached. Amplification is shown also in the receiving consonant branch at 50, in case amplifisonant soundsto their proper amplitude level relative to the; vowels.
The element that tector (36) may be 'with a non-linear input-to-output characteristic.
Instead of using an electromagnetic relay 14 to perform the switching, any other suitable means known to the artl may be used that is capable of exercising the necessary circuit control.
The means and method of the invention is not limited in its use to transmission or receiving systems but may be used to effect separation of vowel components from consonant components for any purpose.
What is claimed is:
l. The method of distinguishing vowel from consonant sound comprising deriving fundamental wave components-from a band of speech frequencies containing harmonics of the fundamental but none of the fundamental frequency component, and utilizing the energy of said derived fundamentalto indicate vowel sounds to the exclusion of consonant sounds.
2. The method of separating vowel from consonant sounds comprising transmitting speech waves over a path normally serving as the con-r sonant path, selecting from the speech waves a frequency band excluding speech fundamental components but including prominent harmonic components of the speech fundamental, deriving from the latter, wave energy of speech fundamental frequency, and utilizing said energy when present to cause the corresponding speech waves to traverse a different path serving as the vowel path.
3. A system for distinguishing vowel from consona'nt sounds comprising means to select from speech waves a frequency band excluding components of fundamental speech frequency but including harmonics of the fundamental, means to derive from such -selected band a component of the speech fundamental, and means controlled by said fundamental sounds to the exclusion of consonants.
a. In combination, a circuit having two branches, means commutating said two branches, said means normally effectively connecting said circuit to one of said branches, means to transmit speech over said circuit, filter means to select a frequency band from said speech excluding the speech fundamental but including harmonics thereof, means to derive a component of speech fundamental from said selected band, and means controlled by said derived fundamental component for operating said commutating means to effectively disconnect said circuit from the first it to the other branch.
level occupied by the other group, whereby both groups of components are transmitted in a band of frequencies less wide than that of the speech components as produced.
6. The method of separating vowel from consonant sounds comprising exploring speech waves to determine the presence of components rich in harmonics of the fundamental speech components as distinguished from components relatively lacklng in such harmonics, and utilizing the energy of JOHN C. STEINBERG.
so derived for indicating vowel