GB2154782A - Method of and apparatus for detecting and analyzing musical phrases which are sung or played - Google Patents

Abstract

The method of and apparatus for detecting and analyzing musical phrases which are played or sung consists in ascertaining at least the fundamental frequency of any note sung or played from out of number of semitones over a given range of frequencies. A plurality of resonators (RZ) are provided, one for each semitone in the range, each including an operational amplifier (K5). Each resonator also includes a peak voltage detector comprising an operational amplifier (K6) and diodes (D5) and (D6), the latter diodes (D6) being connected to a discharge bus. A comparator then cyclically compares voltage outputs from adjacent resonators in order to ascertain which one has an output which has peaked to a maximum. This information is fed to a computer which can then verify whether the musical phrase sung or played is correct. <IMAGE>

Description

SPECIFICATION Method of and apparatus for detecting and analyzing musical phrases which are sung or played The present invention relates to a method of and apparatus for detecting and analyzing musical phrases which are sung or played.

More particularly the method and apparatus are designed to be used with a computer primarily for the purpose of teaching music to pupils or students.

Thus the method and apparatus is required to detect at least the fundamental of any note sung or played from out of a number of semitones over a given range, in order that this information can be presented to a computer in digital form for the purpose of analysis.

Since the advent of the cheap micro-computer, Computer Assisted Learning (CAL) has become well-established in many fields at all levels from pre-school to university. It has, however, as yet had little impact on music teaching apart from some software aimed at teaching elementary harmony and notation.

Its impact on practical music teaching is virtually non-existent. The reason for this is without doubt the lack of a suitable input method for the student to feed back his responses to the computer.

As an example of the problem we are addressing, consider the following. In testing and developing musical ability in young children a short melodic phrase is played by the teacher and the child is asked to sing it back.

The response is then checked for accuracy of pitch and rhythm. Although there is absolutely no problem in having a computer play the original phrase there is no method currently available for a microcomputer which will enable it to accept the sung response and react to it in real time as does the human teacher. The currently available computer inputs, such as the standard typewriter ("OWERTY") keyboard, joysticks, or light pens, are quite useless in this respect.At a more advanced level of music instruction it may be possible for a pupil who has had some instruction in piano or other keyboard instrument to feed his response to the computer via a suitably programmed QWERTY keyboard (although experience has shown this to be very slow, clumsy and error-prone) or, better, by an electronic organ keyboard (and relatively inexpensive versions of such an interface are, or soon will be, available for a variety of micro-computers).

Nevertheless, the ability to translate a tune in one's head, which one will have no difficulty in singing, into a played version on a keyboard requires a musical sophistication far in advance of the level being considered here and even this level of sophistication does not help the non-keyboard player (e.g. singer, violinist, or clarinettist) to input his response.

The present invention relates to a method whereby a micro-computer can accept a musical input through a microphone and determine in real time what note is being played or sung. The computer will be able to store musical phrases so produced as a sequence of notes with their durations, and respond accordingly to the pupil via its CAL software.

The idea of communicating with computers via sound is not in itself new. It is said that the "fifth generation" of computers (when they are available) will be able to respond directly to spoken input, and will therefore no doubt take the matters under discussion here in their stride. However these computers are aimed at industrial robotic applications and will be expensive. It looks like being some considerable time yet before their cost and availability makes them comparable with present-day cheap micro-computers. There are some present-day toy micro-controlled robots which can respond to simple spoken commands like "up", "down", "left", "right", but they generally only respond to the person whose "voice-print" they have been programmed with, and their response time is relatively slow.At a more sophisticated and relevant level there are computersynthesizers like the Fairlight CMI (Computer Musical Instrument). This can accept sound input through a microphone and analyse the harmonic structure in great detail, using Fourier analysis. It can store the structure and later reproduce the sound either "straight" or modified in some way ("bent") at the user's behest. The Fairlight is a splendid instrument, but it costs about 20,000 and is clearly only likely to find application in music studios. Its capabilities are far in excess of what is required here. In any case Fourier analysis is not a particularly fast routine, requiring a lot of calculations, a lot of computer memory and precludes response in real time unless the computer involved is fitted with a hardware Fourier analyser, which is not cheap.

It is therefore an object of the present invention to produce a reasonably cheap means whereby a standard micro-computer can be enabled to accept musical phrases sung or played to it directly without the requirement of any special training on the part of the user. It should cover at least a 3-octave range from G (bottom line of the base clef) i.e. 98 Hz, to G (space above the top line of the treble clef) i.e. 784 Hz which is roughly the range of the human voice, and it should be able to respond in real time i.e. the user should not have to wait a significant length of time after he has finished his input for the computer to deal with it. This effectively precludes techniques like Fourier analysis. The invention relies on the fact that we are dealing with a highly specialized type of sound.It is known that the sound input will have a defin ite, well-defined frequency structure, that the lowest frequency present will be (at least approximately) one of 37 well-defined frequencies (for a 3-octave range, though the principles apply equally to wider ranges), and that the frequencies present will consist of a fundamental plus overtones. It is only required that the fundamental frequency be determined. The requirement that it respond in real time suggests that it should be able to cope with notes which change at a rate of up to say 10 times per second.

One possible approach which would use minimal hardware additions is via digital filtering. A digital filter is an algorithm which produces a series of outputs f, (t = 0, 1,2,.....) from an input sequence d, (t = O, 1,2 ). It can be so arranged that if d, represents a digitized version of a sine wave, f, will also be a (digitized) sine wave of the same frequency, but whose amplitude depends on that frequency-in other words it behaves as a filter and can be arranged to have any of the standard filter characteristics e.g. low-pass, band-pass etc. The turn-over frequencies of such filters are altered by changing the parameters of the algorithm, which can be done by software.To implement such an arrangement it would be necessary to digitize the incoming signal at a rate which depends on the highest pitch it is required to recognize (which would need to be about 2KHz for the range suggested), for say 0.1 second, and store the resuits in memory. This would require 200 bytes of memory if single byte (8 bit) resolution is sufficient, or 400 bytes if it is not. These are not particularly demanding requirements of modern microchip technology. There are a number of fairly cheap analog-to-digital converters available capable of working at well beyond that speed, and the memory requirement could be supplied by all but the very smallest of microcomputers without undue embarrassment. The digital filter would be set up as a narrow band-pass, first at the lowest frequency of interest, and passed over the data series.The amplitude of the steady-state part of the response would be determined and stored (note that as with all filters, the response would include a 'transient'). The filter would then be re-programmed to respond at the next semitone and passed over the series again. This is repeated until a maximum response amplitude has been noted (i.e. the amplitude has started to decrease) or the highest frequency has been tested. the simplest band-pass digital filter requires four multiplications and additions per data point. For 200 data points this is 800 multiplications and additions per frequency tested. Multiplication is a slow process on standard microprocessors roughly 50 cycles on a microsecond processor, even for a single byte precision. Thus it would require about 40 milliseconds per frequency tested, so it would appear to require about 1.8 seconds to test all the frequencies of interest.This is clearly much too slow. It may be possible to reduce this time a little by using fewer data points, but it should be noted that the numbers suggested only cover about 10 complete cycles at the lowest frequency of interest, which is about 100 Hz. Given the fact that the microprocessor will also need time to address the analog-to-digital converter, read in the results, test the amplitudes and check the student's responses, it is clear that something like a 50-fold increase in microprocessor speed is likely to be necessary to enable such an approach to work at the speed required, and even the promised 1 6-bit microcomputers will not run that much faster.

The above difficulties arise from the fact that microprocessors are serial devices. They work fast and can jump around from one job to another and back, but can still only work on one job at a time. The essence of the present invention is to introduce a small but sufficient amount of parallel processing to enable all the frequencies of interest to be recognized simultaneously.

According to a first aspect of the present invention there is provided a method of detecting and analyzing musical phrases which are sung or played, said method including the steps of: (a) detecting the fundamental frequency or two harmonics of the fundamental frequency in order to ascertain the fundamental frequency of the note sung or played; and (b) inputting into a computer in digital form for analysis, information relating to a musical phrase which has been sung or played.

In the case where the method is applicable to the teaching of pupils or students, the computer may additionally be programmed to verify whether or not said musical phrase has been sung or played correctly, by giving a visual or audible output.

According to a second aspect of the present invention there is provided apparatus for detecting and analyzing musical phrases which are sung or played, said apparatus including: (a) a plurality of resonators, whose frequencies are tuned to respective semitones over a range of a given range of frequencies to be sung or played; (b) detector means for detecting that resonator (in the case where a fundamental is detected) or a pair of resonators (in the case where harmonics of the fundamental is detected) which has or have responded most strongly to the frequency sung or played; and (c) means for inputting to a computer in digital form for analysis, information pertaining to the musical phrase which has been sung or played.

Additionally, in the case where it is desired to teach pupils or students the computer may be programmed to derive a visual or audible output which is a verification as to whether or not said musical phrase has been sung or played correctly.

The present invention will now be described in greater detail by way of example with reference to the accompanying drawings wherein: Figure 1A is a circuit diagram of one of the thirty-seven resonators, used to detect the frequencies output from a microphone; Figure 1 B is a graph illustrating the outputs from two adjacent resonators; Figures 2A and 2B of two alternative forms of peak voltage detectors for detecting the output from a resonator shown in Figure 1 A; Figure 3A is a circuit diagram of a combined modified resonator and peak voltage detector; Figure 3B is a graph showing a characteristic of the circuit shown in Figure 3A; Figure 4 is a block circuit diagram of one preferred form of electronic system for use with a computer to detect, analyze and verify musical phrases which are sung or played.

Referring first to Figure 1A, each of the thirty-seven resonators to cover the range of from G(98Hz) to G(784Hz) includes: three operational amplifiers K1, K2 and K3; a resistor R1; variable resistors VR1 and VR2; and capacitors C1, C2 and C3. The output from a standard microphone and amplifier (not shown) used to pick-up sounds which are sung or played, is applied as an input voltage V,, to all thirty-seven resonators arranged in parallel. In each case the input voltage V,, is applied to an input of the first operational amplifier K1 through the variable resistor VR1. The output of the operational amplifier K1 is applied to an input of the second operational amplifier K2 through the resistor R1. The capacitor C2 is connected between this input of the operational amplifier K2 and earth.The output from the second operational amplifier K2 is connected to an input of the third operational amplifier K3 through the capacitor C3. The capacitor C1 provides a feedback path from the output of the operational amplifier K2 to the input of the operational amplifier K1.

The output from each resonator is applied to respective peak voltage detectors. The first type of voltage detector shown in Figure 2A includes: diodes D1 and D2; a resistor R2; and a capacitor C4. The output from the third operational amplifier K3 is applied to the anode of the diode D1. The diode D1 and the resistor R2 are in series between the output of the third operational amplifier K3 and the respective input to a pair of multiplexer units shown in Figure 4. The diode D2 is connected between the input to the multiplexer unit and a discharge bus (shown in Figure 4). The capacitor C4 is connected between said input and earth.

An alternative form of peak voltage detector is shown in Figure 2B, and includes: an operational amplifier K4; diodes D3 and D4; a resistor R3 and a capacitor C5. The output from the third operational amplifier K3 is applied to the non-inverting input of the operational amplifier K4. The diode D3 and the resistor R3 are in series between the output of the operational amplifier K4 and the respective input to a pair of multiplexer units. The diode D4 and the capacitor C5 are connected in similar manner to the diode D2 and capacitor C4 of Figure 2A. A feedback connection is provided between the junction between the diode D3 and the resistor R3 and the inverting input to the operational amplifier K4.

The resonators shown in Figure 1 A are based on a band pass filter design. Whilst this is satisfactory for sine wave inputs, experience has shown it to be less so for real accoustic inputs owing to the strong harmonics which are usually present, particularly on the lower notes. Although more effective filtering of the harmonics is possible by using high-order filters, this would add considerably to the cost.

Accordingly, in a combined modified resonator and peak voltage detector shown in Figure 3A, each of the thirty-seven resonator and detector units includes: two operational amplifiers K5 and K6; resistors R4, R5 and R6; a variable resistor VR3; capacitors C6, C7 and C8; and diodes D5 and D6. The low pass filter comprises the elements: R4, VR3, C6, C7 and K5 and the resonating characteristics thereof is shown in Figure 3B, where the output at the terminal TP connected to the input to the operational amplifier K5 is plotted to a base of frequency. The peak detector comprises the elements: R5, R6, C8, D5 and K6.

Referring now to Figure 4, the electronic system includes: 37 resonators RE numbered 1 to 37; two multiplexers MPX1 and MPX2; flip-flop FFT, FF1 and FF2; comparators COMP1 and COMP2; a quad bilateral analogue switch OBS; counters C01, C02, C03 and C04; inverters INV1 and INV2; AND gates G1, G2, G3 and G4, a clock pulse generator CPG, buffer amplifiers B1 and B2 and an electronic switch SW. The outputs from the thirty-seven resonators RZ are applied to the inputs of the multiplexer units as follows. The first multiplexer MPX1 receives the outputs from the even numbered resonators on its respective inputs 1 to 18.The second multiplexer MPX2 receives the outputs from the odd numbered resonators on its respective inputs 0 to 1 8. The input 0 of multiplexer MPX1 is earthed, and the remaining inputs of both multiplexers MPX1 and MPX2 are connected to a fixed potential Vb.

The discharge bus is connected to earth through the electronic switch SW and controlled from the flip-flop FF2.

It will be appreciated that the specific construction and interconnection of the circuits shown in Figure 4 will depend on the type of computer with which the system is used.

The operation of the apparatus for detecting analyzing and verifying the accuracy of musical phrases which are sung or played, will now be described in greater detail.

The output from the microphone amplifier is applied to the thirty-seven resonatorss RZ in parallel. each resonator RZ is in the form of a resonating low-pass filter, tuned to a respective different semitone. Each resonator RZ will give a response at the input frequency, but those which are tuned to fundamental or harmonic frequencies present in the input signal will give outputs which are greater than the immediate adjacent resonators. Comparison of the outputs from the resonators can be made by the comparator COMP1 with the aid of the multiplexers MPX1 and MPX2. Due allowance is made for the fact that the sounds of interest change, and every such change will induce a transient as well as a steady state effect.The preferred form of resonators RZ used (Figure 3A) are adjusted to have a peak insertion gain of 20db; appropriate initial setting being achieved by the variable resistor VR3 as well as by adjusting the gain of the operational amplifier K5. If the peak insertion gain of the resonator is set too low, the resonator will have poor discrimination between its resonant frequency and frequencies close to this, e.g. a semitone away. If the peak insertion gain of the resonator is set too high, the transient time constant will be long and the circuit will be difficult to adjust and liable to oscillate.

The dynamic equations for such a circuit can be solved and the response to a driving voltage calculated. Figure 1 B shows (graphically) the response (curve Z) of such a circuit to a sine wave at its resonant frequency applied at time = 0, the circuit having previously been at rest. Superimposed on this is the response (curve Y) of the circuit tuned one semitone higher to the same input. It can be seen that the transient lasts effectively for about 10 cycles, though the two responses are pretty well differentiated after 8 cycles. A similar effect will occur when a note ceases or changes in that the previously excited resonator's response will decay in about 8 cycles of its resonant frequency and the newly excited resonator will build up its response in that number of cycles of its resonant frequency.

The lowest practical note for a direct reading of the fundamental in one tenth of a second is therefore about 80 Hz under ideal circumstances (approximately E flat below the bass stave). The worst case would be where one was trying to take readings 0.1 second apart of a pitch of about 100 Hz or below which was changing 10 times a second exactly halfway between reading points. This could give rise to some confusion. However for acoustic reasons (possibly connected with the above) such rapid changes of low pitch notes are unusual if not impossible with the voice or normal acoustic instruments.

In keeping with modern practice an active filter design would be preferred for the resonators rather than the use of bulky and costly inductors. It is important that the resonators have, as far as possible identical characteristics apart from their actual resonating frequencies. Thus it is necessary to be able to trim them for frequency, and insertion gain. It is possible to design a circuit having the appropriate transfer function with a single amplifier. The section comprised of components R4, VR3, C6, C7 and the operational amplifier K5 form a Sullen and Key low-pass filter giving a complex pair of poles. The circuit should be monitored at TP and a signal V, of the appropriate frequency applied.The monitoring at the point TP consists of measuring the peak voltage obtained thereat using a high impedance voltmeter or oscilloscope. The variable resistor VR3 is adjusted to cause the response to peak at this frequency. Once the frequency is set, the insertion gain can be adjusted by altering the gain of the operational amplifier K5 (this is done using a variable resistor in its feedback circuit), and it is set to give a peak to peak voltage at TP 10 times that of the input peak-peak voltage.

The peak voltage is detected by attaching the diode D5 and RC network consisting of the resistors R5, R6 and capacitor C8 to the output of the buffer amplifier K6, with feedback taken from the junction between the diode D5 and the capacitor C8. It will be necessary to arrange that the capacitors C8 can be discharged before each new reading is taken. This can be done by connecting each capacitor C8 to a "discharge bus" via the isolating diodes D6. The discharge bus itself is controlled by an electronic switch SW. It should be noted that if the discharge bus is taken to zero volts at the start of a read cycle, the voltage at the anode of the diode D6 will only fall to 0.7 volts due to the normal diode threshold voltage. In order to ensure that the voltage on the capacitor C8 is quickly restored to a known condition after a discharge, even in the absence of a signal, the A.C. voltage from the microphone amplifier at Vm has a D.C. voltage of 0.7 volts superimposed on it.

This is referred to in the following description as Vb.

Imagine the resonators RZ set out in line in order, lowest pitch on the left. Suppose a signal of fundamental frequency f is applied to all the resonators. For each resonator which is tuned to a frequency below fits response will be less than that of its right-hand neighbour, since the latter's tuning will be closer to f. For resonators tuned to f or above, the reverse will be the case. If the input signal contains some harmonics, similar responses will be observed in the region of 2f, 3f etc. The important point is that by comparing the outputs from adjacent resonators one can determine what frequencies are present, to the nearest semitone, in the input signal. This is done by the comparator COMPI, which is a standard electronic circuit that has two analogue inputs ("inverting" and "non-inverting") and a logic (i.e. digital) output.If the voltage on the non inverting input is higher than that on the inverting input the output is logic 1 and vice-versa. Since comparators can work very fast (a few nano-seconds is usually enough for them to give a response) it is feasible for one comparator chip to do all the comparisons required in a few micro-seconds.

The complete system, as shown in Figure 4, consists of the 37 resonators RZ as described above, each with its peak detector connected via its isolating diode D4 to the discharge bus, which is itself controlled by the transistor switch SW. The outputs of the resonators RZ are connected as shown to the inputs of 2 20way to 1-way analogue multiplexers MPX1 and MPX2. Inputs 1-18 of the multiplexers MPX1 are connected to the even-numbered resonators RZ, input 0 being earthed and input 1 9 being connected to a potential Vb.

Inputs 0-18 of the multiplexer MPX2 are connected to the odd-numbered resonators RZ, input 1 9 being connected to the potential Vb. The multiplexers are controlled by the outputs of 2 5-bit counters C01 and C02 which are themselves clocked from the outputs Q and Q of a toggle flip-flop FFT. The effect of this arrangement is that the counters C01 and C02 each advance on alternate clock pulses fed to the flip-flop FFT. The outputs from the multiplexers MPX1 and MPX2 are fed via a quad bilateral analogue switch QBS, which is set up as a 2-pole double-throw switch by connecting it as shown.The switches are controlled by the outputs of the flip-flop FFT so that when the output Q is high, the output of the multiplexer MPX1 is connected via a buffer amplifier B1 to the inverting input of the comparator COMP1 and the output of the multiplexer MPX2 is connected via a buffer amplifier B2 to the non-inverting input of the comparator COMPi. When the output 0 is low these connections are transposed. The overall effect is that at any time, two adjacent resonators are connected to the comparator COMP1 inputs, the higher-numbered resonator RZ always being connected to the inverting input.

A resistor Rh connected between the output of the comparator COMP1 and its non-inverting input, provides 1 00mV or so of hysteresis to prevent the comparator responding to random noise.

The remaining inputs of the multiplexers MPX1 and MPX2 can conveniently be thought of as connected to "pseudo-resona tors ' numbered 0, 38 and 39 which are connected in accordance with the normal pattern but which give permanently zero or Vb outputs. The D.C. output from the real resonators never drops below the potential Vb, thus resonator 1 is always giving a higher output than resonator 0. After a reset (see later) when the counters C01 and C02 are both at zero, these two resonators are the ones presented to the comparator COMPi. As a result, the output of the comparator COMP1 goes low. Pairs of resonators (1,2 2,3 3,4 etc.) are now compared until a position is found where the higher one is giving the smaller output.

This causes the comparator COMP1 to switch high and indicates that the lower one of the current pair is the peaking resonator. If the comparison process is carried on beyond this point all subsequent comparisons should give the high output until the effect of harmonics begins to be found. With a natural sound which has strong harmonics this can happen fairly quickly. However a single isolated comparison giving a high output at the comparator COMP1 can arise in odd places due to a transient when notes change. It has been found in practice that two adjacent comparisons giving high outputs at the comparator COMP1 is a much more reliable indicator of true pitch than is one. This is dealt with by the components which include a resistor Rt, a capacitor Ct and a comparator COMP2. The type of comparator envisaged for COMP1 has an uncommitted collector output.Consequently, when the output is high, the capacitor Ct starts to charge up through the resistor Rt. The time-constant Rt-Ct and the resistor network around the comparator COMP2 are so arranged that the voltage on the capacitor Ct will not rise far enough during one clock pulse to trigger the comparator COMP2, but will do so during two clock pulses. If the comparator COMP1 goes low after only one clock pulse, the capacitor Ct is discharged again, and the comparator COMP2 is unaffected.

The output of the comparator COMP2 is fed to the reset input of a flip-flop FF1 which is a D-type flip-flop whose D pin is connected to logic 1. The output of the flip-flop FF1 going low closes an AND gate G1 and prevents any further clock pulses reaching the flip-flop FFT or a 6-bit counter C03 from a clock pulse generator CPG. Thus the count in the counter C03 at this point (and until it is reset) is one more than the number of the resonator which is peaking. Pseudo-resonators 38 and 39 ensure that a valid result can be obtained for resonator 37. If there is no signal input, all resonators (except number zero) will be giving the same output (Vb) and the comparators will not change state. This situation is easily recognized by the counter C03 having a count greater than 38.

The clock pulse generator CPG should run at a frequency of about 45KHz. It supplies pulses to the flip4lop FFT and the counter C03 via the gate G1. It also supplies various other control signals via a 10bit counter C04, three AND gates G2-G4 and a D-type flip-flop FF2 which has its D pin connected to logic 1.

These signals are derived as follows. The counter C04 counts up to 562 (bits 2,5,6,10 set). At this point the output of the gate G4 goes high sending a reset signal to the counters C01, C02 and C03 and the flip-flop FFT (this signal line is not shown on Figure 4 for clarity). It is also connected to the reset input of the flip-flop FF2 and, via a short delay (to ensure that the signal exists for long enough to reset the devices it is connected to) to the reset input of the counter C04 itself. Thus almost immediately after the count reaches 562, all the counters are reset to zero and the 0 and Q outputs of the flip-flop FFT are set to zero and one respectively. The Q- output of the flip-flop FF2 is also set to one, and this switches on the discharge bus switch SW.

The reset input on the flip-flop FF1 will go low but irrespective of the state of the Q output, pulses through the gate G1 will be inhibited by virtue of bit 10 of the counter C04 being low. When the count in the counter C04 equals 48 (bits 5 and 6 set, after approx 1 msec) the output of the gate G2 goes high clocking the flip-flop FF2 and switching off the switch SW. When the count equals 512 (bit 10 set, after a further 10 msecs. approximately to allow the capacitors in the peak detectors to recharge) the flip-flop FF1 is clocked high and clock pulses are passed by the gate G1 until either the flip-flop FF1 is reset as described above, or the count reaches 562. When the count reaches 560 the output of the gate G3 goes high sending a "data available" signal to the computer.

Typically, it will be arranged for this to generate an interrupt. The computer then has about 44 jusecs in which to read the contents of the counter C03 before the whole cycle is restarted on the count of 562. Any of these figures are, of course, easily modified to suit circumstances.

The inverters INVI and INV2 are provided for the synchronization of the complete system, since the flip-flops clock on the rising edge of a pulse whilst the counters clock on the falling edge.

In addition to the above described apparatus, the following additional and/or alternative embodiments can readily be achieved.

(a) A recent arrival on the scene of building blocks available to the circuit designer is the "switched capacitor filter" integrated circuit.

These could be used for the resonators. Their advantage lies in the fact that their characteristics are accurately programmable and this could do away with the need to trim the resonators, thus saving production time. At the present time they are moderately expensive, but if their price should fall significantly they could become attractive. Likewise if cheaper and less power-hungry hardware multipliers were to become available, digital filters (as described earlier) could be realized in hardware for the resonators. These too would have the advantage of needing no adjustments.

(b) The essential feature of the invention is the introduction of parallel processing. It may be, however, that having a separate resonator for each note may, in some environments, be a certain amount of overkill. If the resonators shown in Figure 3A are built around transconductance amplifiers, their characteristics become programmable in real time. Thus a possibility would be to have 1 2 resonators, each tuned to a different semitone. They could initially be programmed to cover the lowest octave. If the peak response is not found there they are virtually instantaneously re-programmed (either on-board or by the host computer) to cover the next and higher octaves. It should be noted that the settling time required to allow for transients is inversely proportional to the frequency covered.Hence if the settling time required in the lowest octave is 0.1 second, it is only 0.05 second in the second octave and 0.025 second in the third. Thus the time to explore the 3 octaves is only increased by 75% and 2/3 of the resonators are saved. However much more expensive components are required and the economics would depend on total manufacturing cost as well as timing.

(c) The invention has been described in terms of a 3-octave range. However many popular musical instruments extend beyond this (e.g. violin, flute at the top; cello at the bottom). Extending the range upwards is no problem, it is only necessary to provide appropriate resonators and multiplexers to match.

There could be a potential problem in trying to extend the range downwards. Recall that when a sound changes it takes about 10 cycles for the resonators to settle down. There is therefore little point in trying to read the resonators more frequently than the time of 10 cycles of the lowest frequency covered. If it is required to be able to keep up with music moving at a reasonable pace the figures so far quoted seem to suggest a limitation. There appears however to be a way round this limitation for low notes, which is to make use of the fact that all natural (i.e. non-electronic) low-pitched musical sounds have a rich har monic structure which could be exploited. If instead of storing just the lowest frequency found, the first two peaking resonators are stored (this is easily arranged by duplicating the counter C03 and arranging the gating so that on the first peak found one counter stops counting, but comparisons continue until a second peak is found, which stops the second counter and the comparison procedure), the host computer can then compare the two frequencies. If the ratio is 2:1 then the lower one is the fundamental of the sound being entered. If the ratio is 3:2 then the frequencies observed are the second and third harmonics, and the fundamental is at half the lower frequency. This method should work with all known musical instruments or voices apart from those of the clarinet family, which by their nature give out only odd-numbered harmonics. Thus the first two overtones emitted are the 3rd and 5th harmonics.However the computer software can deal with this just as easily. If it finds the frequency ratio of the two detected frequencies is 5:3 it will know that it is listening to a clarinet (or bass clarinet) and that the fundamental is at one third of the lower frequency. It will be obvious that this method in general offers a potential means of extending the range of the system by an octave or so without providing resonators in that octave and this could therefore possibly be used to reduce the number of resonators required even in a basic 3-octave system, thus reducing the cost. However its practicality is still to be investigated.

(d) If instead of just examining the resonators to determine which one or two lowest frequencies are present, all the frequencies (harmonics) are located and their relative strengths measured (the preferred method for doing this would be via an analog-to-digital converter which would transmit the digital peak output voltage from each peaking resonator to the host computer), it would be possible for the latter to determine the harmonic spectrum of the sound stimulus. Thus the system could be easily extended to be a cheap but effective harmonic analyzer which could have uses in the teaching and study of musical acoustics.

(e) It should be noted that although the above described apparatus is intended for serious purposes, e.g. teaching; it also has considerable recreational and commercial potential. For example, it could be feasible that a computer connected to a synthesizer could capture a sung phrase or tune and play it back in the timbre of any chosen instrument, print it out in conventional music notation, or even harmonize and orchestrate it.

(f) In a further alternative form, instead of the resonators being based on either a band pass filter as shown in Figure IA or a low pass filter as shown in Figure 3A, the resonators can be based on notch filters (anti-resonators).

This type of filter gives minimum rather than maximum response at the tuned frequency.

The principle would be the same except that one would be detecting which resonator had substantially zero response.

(g) Finally, much recent work in computing has been devoted to trying to make computers 'understand' ordinary speech, so that the spoken word can be transcribed by computers, translated into other languages or used to issue commands to computers perhaps to control other devices. Very large computers can now, to some extent, perform these functions but at high cost. Microcomputers can only offer very limited capability with the last of these if the instructions are very simple and the list of possible instructions very short, due to the large amount of processing and memory required by present techniques. The present invention makes it possible for microcomputers to recognize musical phrases or tunes, to which of course, words could be sung.Thus if instructions were to be sun g--each instruction having its own tuns the possibility exists of using microcomputers in these sorts of applications much more effectively. The computer would be recognizing the tunes rather than the words, but there are already precedents for issuing instructions musically-army bugle calls, for example, or the shepherd whistling instructions to his dog (and why should a micro-computer be expected to be any more "intelligent" than a dog?). Furthermore, it is well known that words and music complement one another, a fact exploited by many religious practices etc., where knowing the tune helps one remember the words and vice-versa. Thus it would not be difficult for a potential user to learn a set of tunes which fit the words of his intended set of instructions and to sing his instructions to the computer.

Claims (11)

1. A method of detecting and analyzing musical phrases which are sung or played, said method including the steps of: (a) detecting the fundamental frequency or two harmonics of the fundamental frequency in order to ascertain the fundamental frequency of the note sung or played; and (b) inputting into a computer in digital form for analysis, information relating to a musical phrase which has been sung or played.

2. The method according to claim 1, which includes the additional steps of: (c) pre-programming the computer to verify whether or not said musical phrase has been sung or played correctly; and (d) obtaining a visual or audible output to indicate the result of said verification.

3. The method according to claim 1 or 2, wherein the step of detecting the fundamental frequency includes cyclically comparing the outputs from adjacent resonators over a plurality of resonators tuned to semitones over the range of musical phrases to be sung or played and ascertaining which resonator is providing the greatest peak output voltage.

4. The method according to claim 1 or 2, wherein the step of detecting the fundamental frequency includes comparing the outputs from adjacent anti-resonators over a plurality of anti-resonators tuned to semitones over the range of musical phrases to be sung or played and ascertaining which anti-resonator is pro viding the smallest or substantially zero output voltage.

5. Apparatus for detecting and analyzing musical phrases which are sung or played, said apparatus including: (a) a plurality of resonators, whose frequencies are tuned to respective semitones over a range of a given range of frequencies to be sung or played; (b) detector means for detecting that resonator (in the case where a fundamental is detected) or a pair of resonators (in the case where harmonics of the fundamental is detected) which has responded to the frequency sung or played; and (c) means for inputting to a computer in digital form for analysis, information pertaining to the musical phrase which has been sung or played.

6. Apparatus according to claim 5, including visual or audible output means to indicate the result of a verification that said musical phrase has been correctly sung or played, said verification being obtained from the computer which has been appropriately programmed to effect said verification.

7. Apparatus according to claim 5 or 6, wherein said detector means includes a comparator which cyclically compares the outputs from adjacent resonators to determine the one having the greatest peak output voltage.

8. Apparatus according to claim 7, wherein said resonators are replaced by anti-resonators, and the comparator of said detector means cyclically compares outputs for adjacent anti-resonators to determine the one having the smallest or substantially zero output voltage.

9. Apparatus according to claim 7, wherein each resonator comprises an operational amplifier having an input connected to a microphone via a first resistor and a variable resistor, a first capacitor connected between the output of the operational amplifier and the junction between the first resistor and variable resistor, and a second capacitor between the input to the operational amplifier and earth.

10. Apparatus according to claim 9, wherein said detector means has asssociated with each resonator a peak voltage detector comprising a second operational amplifier whose non-inverting input is connected to the input to the first operational amplifier, a second resistor and a first diode in series between the output from the second operational amplifier and an input to said comparator, and a second diode connected between the input to the comparator and a discharge bus, the inverting input of said second operational amplifier being connected to the junction between said two diodes and constituting a feedback path.

11. Apparatus according to claim 10, wherein said comparator means includes: a pair of analogue multiplexers, the inputs to one of said multiplexers being connected to outputs of even numbered resonators and peak voltage detectors, whilst the inputs to the other of said multiplexers are connected to outputs of odd numbered resonator and peak voltage detectors; a clock pulse generator; a pair of counters for controlling the respective multiplexers; and a toggle flip-flop having Q and Q outputs which clock the respective counters, the arrangement being such that the counters are each advanced on alternate clock pulses fed from the flip-flop.

1 2. The method of detecting and analyzing musical phrases which are sung or played substantially as herein described with reference to the accompanying drawings.

1 3. Apparatus for detecting and analyzing musical phrases which are sung or played constructed substantially as herein described with reference to and as illustrated in Figures 1A, 2A and 4, or Figures 1A, 2B and 4, or Figures 3A and 4 of the accompanying drawings.