JPS5945160B2 - Voice identification method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a voice identification method for differentially controlling mechanical or electronic devices such as powered prosthetic hands and manipulators by encoding voice signals.

Conventionally, this type of method analyzes the frequency of various voice signals from the caller, records each voice signal as a pattern, and then analyzes the frequency of the voice signal uttered by the caller the next time. and compared with the above pattern,
The purpose was to determine what kind of audio signal it is.

However, with this method (the voice signals of large callers vary depending on the sign or time, i.e., the intervals between each syllable are different or the pronunciation is different), it is difficult to compare patterns. In addition, the device was large and the cost was high.This invention was made in view of the above points, and its first purpose was to attach a microphone to the outer wall of the trachea. An object of the present invention is to provide a voice identification method that allows information to be confirmed even in normal vocalization or humming.

A second object of the present invention is to provide a speech identification method that can obtain a high recognition rate because the identification is performed by classifying the number of syllables and the pitch change pattern of the syllables.

Next, one embodiment of the present invention will be described with reference to the drawings.

FIG. 1 is an overall block diagram. A is a voice detection circuit that converts the voice signal from the caller into a pulse signal, 1 is a microphone installed on the outer wall of the voice caller's trachea, and 2 is a filter that passes 50 to 250 Hz. , the pitch component, which is the fundamental frequency of voice vibration and is a parameter representing the scale at the time of vocalization, is extracted. Note that this filter 2
also includes amplifiers. Reference numeral 3 denotes a Schmitt trigger circuit that converts the sine wave extracted by the filter 2 into a rectangular wave and further removes waves other than pitch components using hysteresis characteristics.

B is a count circuit for measuring the period of the rectangular wave output from the Schmitt Trigger circuit 3 using a selected clock, and 4 is a gate for controlling the input of the pulse signal from the Schmitt trigger circuit 3 and controlling the selection of the clock. The circuit 5 is a counter that counts the clock pulses from the clock pulse generating circuit 6 during the signal from the audio detection circuit A, and the numeral 6 is a clock pulse generating circuit that generates clock pulses of, for example, 10 KHz and 300 Hz.

C is a representative value determining circuit that stores each clock pulse number from the counting circuit B and determines a representative value that is an intermediate value; 7 is a latch circuit that stores each clock pulse number;
A comparator 8 compares the number of clocks from each latch circuit 7 and determines a representative value which is an intermediate value. D is a change determination circuit that stores the representative value of each syllable and detects the pitch of each syllable; 9 is a latch circuit that stores the representative value of each syllable;
Reference numeral 10 denotes a comparator which compares the number of clocks from each latch circuit 9 and detects a change L in the number of clocks, that is, the height of a syllable.

E is a syllable break processing circuit that detects syllable breaks using the output from the latch circuit 7 of the representative value determining circuit C, and 11 is a syllable break processing circuit that detects syllable breaks using the output from the latch circuit 7 of the representative value determination circuit C. It is a comparator for comparison. Reference numeral F denotes a system control section which applies pulse signals to each of the above-mentioned programs A to E to perform sequence control.The details will be explained below with reference to FIG. 2 and timing charts. As shown in the speech input envelope in Figure 7, three syllables (for example, ``tsu'' and ``ka'')
When the caller utters "me"), microphone 1 captures this information on the outer wall of the trachea (the part directly below the Adam's apple and is relatively unaffected by high-frequency components and vocal tract characteristics), and captures this information from 50 to A filter 2 that passes 250 Hz extracts the pitch component and amplifies it.

The extracted sine wave is transformed into a rectangular wave by the Schmitt trigger circuit 3 at the next stage. Note that if the caller is a woman, the filter 2 needs to be shifted to the high frequency side. Also, the above rectangular wave is generated due to the hysteresis characteristic of the Schmitt trigger circuit 3.
Furthermore, waves other than the pitch component are removed. This rectangular wave is the a waveform. The entire sequence will now be explained with reference to FIG.

First, O] is an input waiting cycle, in which all circuits are reset to their initial states. 1, C51, (9) is an initial period removal cycle that removes the unstable portion of the pulse signal (in this embodiment, the first pulse) at the rising edge of each syllable.

[Oi, (6), [10] is a cycle that determines the pulse signal to be obtained in each syllable, and in this example, three pulse signals from the 2nd pulse to the 4th pulse of each syllable are measured. There is. 3, [71, [11] are cycles for determining one representative pulse from each of the above three pulse signals, and in this example, the intermediate value is detected, but the average value, the least squares value, the maximum Value, minimum value, nth largest value, etc. can be selected as appropriate.

That is, it is sufficient to detect one pulse signal from three pulse signals in each syllable under the same conditions. [4),8
, [12] is a cycle for determining the break between each syllable. [13] is a cycle for determining the high-low pattern of the three representative value pulse signals, and based on this pattern, information from the caller is determined and, for example, a prosthetic hand starts a "grabbing" action. In addition to controlling prosthetic hands using this pattern, other applications include machine tool control, type control, bookkeeping, and door opening/closing control. The above is a cycle in which sequence control is performed using one piece of information, and after the [0'] input waiting cycle, information judgment is performed again. The above cycle will be explained in more detail.

When the power switch is turned on now, a set pulse is output from the power-on reset circuit 12a through the 0R circuit 12b.
The voltage applied to the bit counter 12c resets the counter 12c and also resets the circuit control section 12d.

That is, the 4-bit counter 12c sequentially instructs the above-mentioned 13 cycles, and is brought into state 6 by the above-mentioned set pulse. Then, the circuit control section 12d that has been set first controls C, d, e, j, j',
Send out pulse signals of j〃, 0, c/', ♂, l. The pulse c is inputted to the input terminal of the flip-flop circuit 4a of the gate circuit 4 in FIG. 2, and outputted from the output terminal Q. This is to cut off audio input until the entire circuit is in the set state. That is, since no output is sent from the output terminal Q side, the AND circuit 4b is closed and does not allow the pulse (hereinafter referred to as an audio pulse signal) from the output trigger circuit 3 to pass therethrough. In addition, the pulse d resets the 8-bit counter 5 or the flip-flop circuit 4c, and when an audio pulse signal is input, output is output from the output terminal Q, and the frequency of 10KHz or 300kHz, which will be described later, is output.
The Hz clock is input to the counter 5 via the AND circuit 4d. Furthermore, the pulse e sets the flip-flop circuit 4e and outputs an output from the output terminal Q at 10KH.
The AND circuit 4f for passing the clock pulse from the clock pulse generating circuit 6a of clock z is placed in a standby state.
Here, since the AND circuit 4d is in a standby state, the 10 KHz clock pulse from the clock pulse generating circuit 6a is input to the 8-bit counter 5 via the AND circuit 4f10R circuit 4g, and the counter 5 is counting. Also pulses J, j', Jl, l and 0,0', cl'
are 8 bit latch groups 7a, 7b, 7c, 7d and 9a,
9b and 9c are cleared to enter the standby state. Next, the circuit control section 12d sends out a pulse b, so the flip-flop circuit 4a is set and an output is output from the output terminal Q and applied to the AND circuit 4b and the flip-flop circuit 4cf) and the D terminal.

In this state, the first pulse a of the audio pulse signal is AN
When input to the D circuit 4b, the pulse a is passed through and applied to the T terminal of the flip-flop circuit 4c (7). At this point, at the rising edge of the pulse, the output terminal switches to the Q side, so a pulse of g is sent out (FIG. 9). At the same time, an increment pulse is applied to the 4-bit counter 12c, and the cycle progresses from the O cycle to the 1 cycle. Then, the pulse d is again sent out from the circuit control section 12d and the 8-bit counter 5 is reset. The flip-flop circuit 4c is also reset, and the output terminal is switched to the Q side. Therefore, the 8-bit counter 5 counts again, but since this first pulse is removed, even if pulse g is generated at the rising edge of the second pulse, the output of the 8-bit counter 5 is not sent to the next stage and is counted again. Reset by pulse d. Then, after the second pulse is input, that is, after the rise time has elapsed, the output terminal becomes Q again, so the 8-bit counter 5 becomes 10.
At the beginning of counting KHz clock pulses, an increment pulse is applied to the 4-bit counter 12c from the circuit control unit 12d, and the third pulse of the audio is inputted as the cycle progresses from the 1st cycle to the 2nd cycle. When g occurs, a pulse i is sent from the circuit starvation section 12d, and the latch 7a (7) StrO in FIG.
It is input to the be end.

Therefore, the count signal from the 8-bit counter 5 is applied to the latch 7.
It is stored in a. Further, since the pulse d is sent out after the pulse i, counting starts at the same time as the 8-bit counter 5 is reset again, and counting continues until the next fourth audio pulse is generated. Then, at the rising edge of the fourth audio pulse, a fourth pulse g is generated, and accordingly a pulse il is generated, and by the same operation, the number of clock pulses corresponding to the third audio pulse is set in the latch 7b, and the number of clock pulses corresponding to the third audio pulse is set to the latch 7b. The corresponding number of clock pulses is stored in each latch 7c. The fifth pulse d which then occurs causes an incremental pulse to proceed from the second cycle to the third cycle. In this third cycle, the outputs (temporarily A, B, and C) of the latch groups 7a, 7b, and 7c obtained in the second cycle are determined, and their intermediate values are determined.

This is performed in the 8-bit comparator group 8a as shown in Table 1 of the appendix. That is, any one of the outputs A, B, and C is sent out from the data multiplexer 8b whose gate is opened by the pulse m from the circuit control section 12d, and the output from the data multiplexer 8b is also output by the pulse n from the circuit control section 12d. It is stored in the bit latch 9a. Furthermore, since pulse f is also sent out at the same time as pulse n, the flip-flop circuit 4e in FIG.
The z clock pulse is sent to AND circuit 4h, 0R circuit 4g, .
The signal is applied to an 8-bit counter 5 via an AND circuit 4d.

After the pulse m, an increment pulse is applied from the circuit controller 12d to the 4-bit counter 12c, and the cycle progresses from the 3rd cycle to the 4th cycle α.This fourth cycle is for detecting the break between each syllable. Similarly to the above, the 8-bit counter 5 counts 300 Hz clock pulses for each pulse d, and receives the pulse k from the circuit control section 12d.
Each signal is stored in the latch 7d and sent to one input terminal of the comparator circuit 11a.

A signal suitable for dividing syllables is applied to the other input terminal of this comparison circuit 11a. Here, the output from the latch 7d is set as A, and the comparison reference output is set as B, and when A≧B, the comparison circuit 11a outputs the output from the latch 7d, which is predetermined to be suitable for dividing syllables. When it is larger, the syllable break pulse s is used, and vice versa A<B
When , a voiced pulse t is sent out. Note that 11b and 11c are AND circuits for obtaining the above-mentioned signal s or t using the timing pulse r from the circuit control section 12d. When the output from the comparator circuit 11a is the voiced pulse t, the above operation is repeated until the syllable dividing pulse s appears, the first pulse of the next syllable is detected, and the pulse s is detected. In this case, pulses E, l, j, j', and jl are sent out from the 4-bit counter 12c.Therefore, the flip-flop circuit 4e in FIG. The 10 KHz clock pulses are counted.Also, the latches 7a-7d in FIG. Proceed to the 5th cycle.The operations from the 5th cycle to the 12th cycle are the same as those described above, so the explanation will be omitted.In addition, the latch 9b in Fig. 4 has the intermediate value of the second syllable, and the latch 9c has the middle value of the 3rd syllable. The intermediate value of is stored.

Furthermore, when h, which is a syllable dividing pulse in the third syllable, is detected, the circuit control unit 12d sends a pulse c as the end of the word, and cuts off the input of the audio signal. Here, the output of each latch group 9a to 9c is added to the adder 1 by adding the upper 4 bits.
It is input to 0a to 10c and added to the original 8 bits, giving a range to the intermediate value of each syllable.

That is, since there are changes in voice, even if the latches 9a and 9b are equal sounds, they do not necessarily match. For this reason, if there is a difference in pitch in vocalization, it is necessary to create a difference of one or more notes. and the first comparator group 10
At d, the signal (1) from the latch 9a, the signal (2) from the adder 10a, and the signal 0 from the latch 9b] and the adder 10
b and the second comparator group 10.
At e, signal 1 from latch 9b is compared with signal (2) from adder 10b, and signal 3 from latch 9c and signal 4 from adder 10c are compared. As a result, 2-bit data output as shown in Table 2 is sent out. For example, if the pitch changes in a word consisting of three syllables as shown in Figure 12A, 1
Outputs of 1 and 01 are output, and when the output changes as shown in FIG. 12B, outputs of 10 and 11 are output. There are a total of nine combinations. Here, the pulse p from the circuit control section 12d is stored in the 4-bit latch 10f, and controls a controller, for example, a motor for "grasping" in a prosthetic hand. Furthermore, after the work is completed, the circuit controller 12d sends out a pulse signal q to clear the 4-bit latch 10f, and then sends out pulses E, l, j, j', Jl, O, O', ol, It returns to the initial state. In the above example, 3
Although recognition of words consisting of syllables has been described, recognition of two and one syllables is also easy.

This will be explained below with reference to FIG. If only one syllable is input now, the process will proceed to the third cycle as described above, and the representative value will be determined and held in the latch 9a.
The clock pulse generation circuit 6b is switched, and the 8-bit counter 5 counts 300 Hz clock pulses. Since the second syllable does not appear here, the 8-bit counter 5 overflows and sends out the pulse h, and the circuit controller 12d
A pulse C is sent out to reset the flip-flop circuit 4a and close the input. Immediately thereafter, pulse q is sent out to clear the 4-bit latch 10f, and pulses U and v are sent out from the circuit controller 12d. Therefore, the inverter circuits 10g and 10h invert the outputs, and the outputs from the comparator groups 10d and 10e are sent to AND circuits 10i and 10.
Press j to stop. This results in a 4-bit latch 10
When a pulse p is input from f, outputs of 00,00 are output as shown in FIG. 12C indicating the end. Next, if the second syllable is input and the third syllable is not input, the process continues up to the seventh cycle described above. That is, the representative values of the first and second syllables are held in the latches 9a and 9b, respectively, and the 8-bit counter 5 counts 300 Hz clock pulses. Since the third syllable does not appear here, the 8-bit counter 5 overflows and pulse h is sent out, so that flip-flop circuit 4a is reset by pulse c. Then, the 4-bit latch 10f is cleared by the pulse q, and only the pulse v is sent out from the circuit control section 12d. As a result, when pulse p is input, only the output of comparator group 10d is input from AND circuit 101 to 4-bit latch 10f. This input is, for example, 11,00 as shown in Figure 12D. Note that even in the case of three or more syllables, recognition is possible by applying the above-described principle.

Furthermore, it is of course possible that the discrimination processing in the above embodiments can be performed by a microcomputer. Furthermore, instead of limiting the classification of change judgment to three (in the example, 2 indicates whether the syllable is above, below, or the same as the previous syllable), by changing the change judgment circuit D, it is possible to detect vertical changes. It can also be subdivided. As described above, this invention attaches a microphone to the outer wall of the trachea where there is little noise, detects the voice uttered by dividing it into moras, and excludes the waveform of the initial few cycles that are not stable as voice, and detects the voice that is not stable as voice. Since the method picks up the period, determines a representative value from among them, and controls the control equipment by classifying patterns based on the height changes of the multiple representative values obtained in this way, it is possible to obtain a high recognition rate. In addition, it has the effect that information can be confirmed even if the sound is normal or humming.

[Brief explanation of drawings]

The figures show an embodiment of the apparatus used in the voice identification method according to the present invention, in which Fig. 1 is an overall block diagram, Fig. 2 is a circuit diagram of the counter circuit section, and Fig. 3 is a circuit diagram of the representative value determining circuit section. 4 is a circuit diagram of the change determination circuit section, FIG. 5 is a circuit diagram of the syllable break processing circuit section, FIG. 6 is a circuit diagram of the system control section, and FIGS. 7 to 11 are timing chart diagrams. 12th A
-D is a diagram showing changes in syllables, and FIG. 13 is a circuit diagram showing another embodiment of FIG. 4.

Claims (1)

[Claims] 1. A means for detecting the sound generated by dividing each syllable with a microphone on the outer wall of the trachea, a means for excluding the initial few cycles of the sound in each section, and a means for determining the division of the syllable.
means for determining a representative value from the measured values of the remaining several periods separated and excluded; and means for determining whether or not there is a detection signal from the microphone within a predetermined time from the time of the separation. , If there is no detection signal within the discrimination time, the pattern is classified based on the representative value, and if a detection signal is present, the representative value is obtained by the same means as described above, and thereafter, while the detection signal may exist, the pattern is classified using the representative value. A voice identification method comprising means for determining a representative value by the same means as described above, comparing each representative value with the previous representative value to determine the pitch of the sound, and classifying the pattern.