JPS6240717B2 - - Google Patents

Description

[Detailed description of the invention]

The present invention relates to a voice generation control device suitable for vending machines, voice guidance devices, etc., and in particular, it selects words suitable for the situation from among various prepared words depending on the input/output situation of the equipment to which it is applied. The present invention relates to a sound generation control device that immediately outputs sound. Recently, as part of the sales promotion of vending machines, there has been a demand for ``Shiyaberu vending machines'' equipped with a voice generator. If you want to generate audio mechanically, the easiest way is to record the audio signal you want to generate on a magnetic tape, disk, drum, etc., and then play it back when needed to output the audio from the speaker. There is a method to generate it. However, this method suffers from wear of the magnetic head, tape stretching, deterioration of sound quality due to wear, and instantaneous song selection (starting the audio).
This method is not suitable for devices such as vending machines that need to output audio instantly in response to changes in external conditions. Therefore, a voice synthesis method has been proposed that combines a microcomputer and a semiconductor memory, digitizes and stores an audio signal, and restores the original audio signal during playback. Typical examples of this method include a waveform reproduction method and an analysis/synthesis method. The waveform reproduction method applies various modulations to the digitized audio signal, stores it in memory, and demodulates it during playback.A typical example is an adaptive differential pulse code that adaptively changes the quantization width. A modulation method (ADPCM) and a constant amplitude step amount Δ (delta) are determined, and a delta modulation method (DM) or adaptive delta is used to encode the residual signal between the previous audio signal and the current audio signal. Modulation method (ADM)
There is. The analysis and synthesis method was developed with the aim of generating vocalizations using even less memory capacity than the waveform reproduction method, and it uses characteristic parameters included in the waveform of the audio signal, such as mouth movements during vocalization, voiced and unvoiced sounds. Only data such as distinctions is extracted and stored, and speech is synthesized based on that data. A typical example is the partial autocorrelation coefficient method (PARCOR). In addition, a lawful synthesis method has been proposed that attempts to elucidate intonation, accent, and other algorithms in human speech and synthesize speech based on character sequence input, but it is difficult to synthesize natural speech. It has many problems and has not been put into practical use. However, in conventional speech synthesis devices such as vending machines that apply these speech generation methods, when the speech generation output is activated, the next speech request is not made until the generation of a series of prepared words is completed. Even if a voice request occurs, it is not possible to immediately switch to the voice request and output it as voice. As described above, there is a problem in that not only is it not possible to output an appropriate message immediately at the time of a voice request, but also it is not possible to take appropriate voice action when a plurality of voice requests occur during voice generation. This problem is particularly important for voice generation control devices such as vending machines, which are expected to be able to quickly respond to changes in various external conditions from time to time and quickly output voice in response to requests according to the current situation. This is a serious drawback. The above-mentioned conventional problems will be explained in more detail with reference to an example of a vending machine below.

【table】

[Table] Let us consider various operating modes of a vending machine that generates the words shown in Table 1. (b) Insert money while pressing the product selection switch. At that time, for example, insert a 100 yen coin while holding down the 50 yen product selection button. The voice from the vending machine executes the steps described above, starting with ``Irritation'' and then proceeding. However, in the case of selling products with a short selling time (e.g. bottles, cans), the customer has already sold the product and has to take out the product and stand up from the vending machine while the customer is talking about "Irritation" in the procedure. I'll leave. Therefore, even after the customer has left, the vending machine will continue to say the following instructions: ``Please take what you like'' and ``Thank you for your continued support.'' (b) The same thing as (b) can occur even if you insert money very quickly, press a button, and take out an item. (c) If you select a product that can be sold after pressing the button several times on a sold-out product or a product that cannot be purchased due to insufficient amount of money, you will hear the voice prompts ``I don't have enough money'' in the step and ``Sold out'' in the step. However, if you press the button too quickly, you will not be able to output an accurate message immediately, such as outputting the procedure when it should have been. If the customer performs an operation that requires the output of a new voice earlier than the time when the voice is generated, a situation will arise in which an accurate message will not be output in response to the actual operation. An object of the present invention is to eliminate the above-mentioned drawbacks, and when a new voice request is made during voice output, the currently output voice is interrupted in the middle or the speaking speed is increased to quickly output the necessary words. An object of the present invention is to provide a high-performance voice generation control device that performs the following functions. That is, the present invention provides a voice generation control device that selects a requested word from a plurality of stored words and outputs it as a voice, in which the word is stored independently for each clause or for each predetermined number of words. death,
A voice production table including an interrupt flag and a speedup flag for each clause or a predetermined number of words; It is characterized by having a control means capable of outputting the new word as audio by referring to both flags and interrupting the audio output of the word currently being output, or speeding up the audio output midway. That is. Hereinafter, the present invention will be explained in detail with reference to the drawings. FIG. 1 shows an example of the configuration of a vending machine system to which the present invention is applied, where 1 is a coin mechanism unit, 2 is a vending control circuit, 3 is a selection button, 4 is a product delivery mechanism, and 5 is a voice generation control. Device,
6 is a PARCOR voice synthesizer, 7 is an audio data ROM (read-only memory), 8 is an amplifier, 9
1 is a speech synthesis section including the above-mentioned parts 6 to 8, and 10 is a speaker. The speech synthesis section 9 has a 3-chip LSI (Large Scale Integrated Circuit) configuration, and can generate a plurality of predetermined types of words (voice messages) in a female or male voice. In addition to the normal generation function, the voice generation control device 5 also has a voice rate switching function and a voice termination function, which will be described later, so that detailed voice information can be provided in accordance with the operating status of the vending machine. First, when a purchaser stands in front of a vending machine and inserts coins or paper, a coin insertion signal is sent from a coin mechanism unit 1, which is usually called a coin pick, to a vending control circuit 2 to a voice generation control device of a voice synthesis section 9. Sent on 5th. Sound generation control device 5
analyzes the received signal, extracts audio data from the audio data ROM 7, and supplies this data to the audio synthesizer 6. The speech synthesizer 6 reproduces the speech signal by the reverse process of the analysis, and this signal is amplified by the amplifier 8 and output from the speaker 10 with the content, for example, "I'm sorry, this is Fuji Electric." generates a sound. Next, the coin mechanism unit 1 compares and calculates the sales set price and the input amount, and when it is possible to sell, sends a sales possible signal to the voice generation control device 5 via the sales control circuit 2. This signal is detected by the control device 5, and a voice is generated from the speaker 10 via the voice synthesizer 6 or the like, saying, for example, "Please press the button of your choice." Next, when the purchaser presses the product selection button 3 and the product is taken out by the product delivery mechanism 4, a sales start signal is supplied from the sales control circuit 2 to the coin mechanism unit 1 and the sound generation control device 5, and the above-mentioned Similarly, based on the audio data taken out from the audio data ROM 7, the speaker 10 is made to say, for example, "Thank you for your continued support." The contents of other voice messages uttered from the speaker 10 are almost the same as those in Table 1 described above. Furthermore, if a bend roulette (not shown) is provided as a bonus device, a winning signal from the bend roulette can be detected and a sound saying ``It's a win'' can be generated by the same means as described above. FIG. 2 shows an example of the configuration of the speech synthesizer 6 shown in FIG. Extract the feature parameters necessary for speech synthesis,
This parameter is supplied to the parameter interpolation circuit. The parameter is a time-domain coefficient k that represents information on the audio frequency spectrum, which is the resonance characteristic of the vocal tract.
_i (1≦i≦p), the loudness (amplitude) of the voice, the frequency of the voice in the voice (pitch period), the presence of voice/
It consists of approximately 150 bits that indicate the distinction between unvoiced sounds and unvoiced sounds.
The parameter interpolation circuit is for interpolating every 2.5ms between frame periods with intervals of 10 to 25ms.
Based on the parameters sent from the parameter conversion ROM, in the case of voiced sound, the impulse signal of the parameter is supplied to the digital filter, and in the case of unvoiced sound, white noise is supplied to the digital filter via the sound source circuit which uses white noise as the sound source δ(n). Feed the signal to a digital filter. The digital filter is a circuit that simulates the human vocal mechanism, and consists of a pipeline multiplier, a stack of k _i parameters, an adder/subtractor, and a shift register. The output signal from the digital filter is passed through a digital-to-analog (DA) conversion circuit to synthesize sound, and the synthesized sound is generated by a speaker 10 via an amplifier 8 as a sound very close to the real voice. Although the speech synthesizer 6 using the PARCOR method described above is relatively superior in terms of speaking time/memory capacity, sound quality, and economical efficiency, the speech synthesizer to which the present invention is applied uses linear predictive coding (LPC). by
Of course, the present invention is not limited to the PARCOR system, and may be a speech synthesizer using other systems, such as a line spectrum pair (LSP) system. FIG. 3 shows a read control unit of audio data read from the audio data ROM 7 of FIG. 1,
Here, A, B, and C are independent words (audio data), and only three are selected from a plurality of words as an example. As shown in the diagram, sentences are separated by natural pronunciation when reading, such as A's ``iratsushi ya imase,''B's ``Fuji Electric,'' and C's ``desimasu''. The speech generation control device 5 performs readout control for each clause, which is the smallest unit (see Fig. 4A).
-C), A, B, and C are connected in this order to synthesize the voices, and a series of messages such as "I'm sorry, I'm Fuji Electric" are uttered from the speaker 10. In other words, the output message is converted into audio data in units of phrases A, B, C, etc. as described later.
Based on the selection information stored in the sound generation table of ROM 7 and supplied from control device 5, A,
Read B, C... independently and calculate the amount of data.
Read at different speeds such as 2.4K, 4.8K, 9.6K bits/second, etc., and stop reading independently. For this reason,
The words A, B, C, etc., which are made independent for each phrase, can be read and stopped for each A, B, C, and so on, and the utterance speed can be varied independently. 4A to 4C illustrate an example of the speech generation control procedure according to the present invention using clauses A, B, and C of FIG. 3. Here, the widths of A, B, and C indicate the time required for utterance. Under normal conditions, when the purchaser presses the buttons on the vending machine at prescheduled intervals, there is no problem in concentrating on uttering a series of messages.
As shown in FIG. A, the control device 5 controls the words A, B, C, etc. to be uttered at a preset normal speaking speed. Next, if there is a request to utter new words Y and Z while the predetermined messages A, B, and C are being uttered, depending on the content of the message and the input/output status of the vending machine, As shown in Figure 4, the speed of speech is increased in the middle of the message being uttered, or as shown in Figure 4C, the following words are interrupted or omitted in the middle of the message being uttered, and a new request is made. The control device 5 controls the messages Y and Z to be uttered promptly. For example, as shown in Figure 4B, when A is in the middle of uttering ``I'm irritated'', Y's request for ``I don't have money'' and Z's request to say ``I don't have enough'' is received. , B's ``Fuji Electric'' and C's ``desimasu'' are uttered at a speed of about 20 to 30% faster than the normal speed. Such variable control of the speech rate cannot be omitted because omitting the latter half of the message series A, B, and C would make the meaning unclear, but it is suitable when the meaning will not become unclear even if the speech rate is slowed down. In this embodiment, speech synthesis is controlled in units of clauses, such as A, B, and C. However, when controlling speech synthesis in units of words, it is possible to It is possible to speed up the speaking speed from the beginning (in the middle of the speech). On the other hand, in the case of Fig. 4C, for example, when Y and Z request to speak a new message while A is saying "Irasu shi ya imase", after A's "Irasu shi ya imase", Suitable when words B and C are optional words. In this case, when the utterance of A is finished, the utterances of B and C are stopped, and the newly uttered messages of Y and Z are uttered. Generally speaking, when a machine, such as a vending machine, is operated by a person, if the person operates the machine more quickly than usual, the machine tries to output the words to be uttered one after another. At that time, the word to be uttered in response to the most recent request should be output immediately, but in conventional voice generation control methods, new words are not output until all the words for which the utterance was previously requested have been uttered. There was an inconvenience that could not have occurred. In the present invention, as described above, it is possible to speed up the speech rate or stop speech midway to quickly produce a new word. For this reason, it is possible to expand the chances of uttering the message, and to output an appropriate message immediately without missing the opportunity to utter the message. FIG. 5 is a flowchart further illustrating the control procedure of FIGS. 4A to 4C. If the voice generation control device 5 is waiting for a voice request, the control is performed in step 21→
It loops like 22 → 21 → 22... Here, if there is a voice request, step 23 to step 2
Proceed to steps 4 and 25 for voice production preparation and voice production processing.
The words selected in steps 24 and 25 are searched for in the table 41 of FIG. 6, and preparation for utterance, utterance start procedure, etc. are performed based on this utterance data. While step 25 is being processed, step 23 is being uttered. Step 2
When the voice production process in step 5 is completed, it is determined in step 26 whether or not a voice request is being waited for. If the voice request is being waited for, control returns to step 21, and steps 21 → 22 → 21 → are repeated until a new voice request is received. 22
...is running in a waiting loop. On the other hand, if a new utterance request is received before the utterance of the message by the voice utterance process in step 25 is completed, the utterance will be interrupted in step 23, so the process will proceed to step 27 and the words (phrases) currently being output will be interrupted in step 23. )
determine whether it can be interrupted. If it is possible to interrupt the speech, proceed to step 28, the speech interruption process, to interrupt the currently outputted word or the subsequent word,
Returning to step 21, the newly requested words are uttered in steps 24 and 25. If it is determined in step 27 that the words currently being output as audio are words that cannot be interrupted, whether or not it is okay to increase the speech rate (speaking speed) of the words currently being output as audio is determined. If it is possible to speed up the speaking speed, the speed-up process in step 30 speeds up the speech generation speed of the word currently being output or its successor, and the speech output ends. Return to step 21 and step 24 and 2
5 generates the new requested word. However, if it is determined in step 29 that it is impossible to speed up the speaking speed, the control returns to step 21, and immediately after the voice output of the series of messages currently being output is finished, step 24 and step Go to step 25 and say the new requested word. Also, if there is a new request while the newly requested word is being uttered, the utterance interruption process in step 28 or the speed-up process in step 30 is performed based on the judgment in step 27 and step 29, so that the opportunity to utter the word is not missed. It is possible to appropriately output a voice message appropriate to the input/output status of the machine (vending machine) at a timing close to the time of request. Note that when the vending operation of the vending machine is completed and there is no longer a need to wait for a request to speak a message, control is returned to the main program (not shown) from step 26. FIG. 6 shows an example of the voice utterance table 41 of the voice data ROM 7 in FIG. 43 analyzes the pre-selected message and creates phrases A, B, C, D...
This is message data that is made independent for each message and connected, and 44 is the END code that indicates the end of the utterance.
The table numbers 42 are assigned corresponding to individual message data 43 selected in advance, and A, B, and C (indicated by diagonal lines) of the message data 43 indicate phrases commonly used in multiple messages. There is. The voice generation control device 7 receives selection information 45 that is supplied according to the external situation (input/output situation) of the vending machine.
Based on the required words A, B, C...
are selected and extracted using the table number 42, and these words A, B, C, . . . are connected to generate a single message. FIG. 7 shows in detail the message data 43 whose table number 42 is 0 as an example of the message data 43 in the voice utterance table 41 of FIG. Here, 43a to 43d are voice data for voice production, respectively, and each voice data 43a to 4
3d each has one clause A or B, C, D and flags F ₁ and F ₂ each consisting of one bit. The flag _F1 is an interruption permission flag used to determine whether it is okay to interrupt words A, B, C, or D of the clause in the utterance data to which it belongs, and the flag _F2 is an interruption permission flag used to determine whether it is okay to interrupt words A, B, C, or D in the clause in the utterance data to which it belongs. Word A or B of the clause,
This flag is used to determine whether or not the audio output speed of C and D can be increased. Both flags F ₁ and F ₂ are set in table 41 in advance. X ₁ to _{X 3} are each utterance data 43
The words A to D of each clause are provided between a to 43d.
It is a word spacing adjustment timer that controls the time to connect the words.
Set the time so that the words you say sound natural. The END code 44 is a special code set at the end of the table data, and the control device 7 determines that the utterance has ended when the END code 44 is input. Flag F ₁ is used in step 27 of FIG. 5, and flag F ₂ is used in step 29 of the same figure. for example,
If there is a new utterance request while the word A of the first utterance data 43a is being uttered, a flag attached to the word A is set by the control device 7 in step 27.
Referring to F ₁ , if it is interruptible, for example "1", the utterance of word A is immediately interrupted in step 28, and the next newly requested word is uttered in step 24.
Then, the table number is extracted from the table number 42, and the process proceeds to step 25, the voice production process. On the other hand, flag F ₁
If, for example, "0" is set to enable interruption, the controller 7 then sets the flag F ₂ in step 29.
Refer to , and if it has a flag of "1" that can speed up the audio output, for example,
In step 30, the voice pronunciation speed of words subsequent to word A is sped up. At this time, each clause A to D
Based on the set times of the word spacing adjustment timers _X1 to _X3 , the time for connecting the words is adjusted so as to be relatively shortened as the speaking speed increases. Furthermore, the flag F ₂
By using a plurality of determination data, it is possible to change the audio output speed in multiple steps. For example, if you express it in hexadecimal, you can get up to 16 variations. In this embodiment, flags F ₁ and F ₂ are referenced for each word in the clause, so even if word A cannot be interrupted, if any of words B, C, or D is interruptible, It is possible to stop the utterance from the word when the interrupt flag data of the interrupt flag _F1 is found and proceed to utter the new word. Also, for the same reason, the words B, C, and D that follow word A
It is possible to speed up all of the words or say only a part of them. As explained above, according to the present invention, predetermined message data is independently combined for each phrase or for each predetermined number of words, and flags indicating whether or not to interrupt and whether to speed up are flagged for each independent data. A speech table with data, and when a new speech request is received while a requested message is being uttered, these flag data are referenced to interrupt the voice output of the series of messages being output or to change the speaking speed. A control means is provided to perform control so as to speed up the message output, thereby outputting an effective message at the point when the voice is desired to be output or with a short waiting time, and giving appropriate instructions to the person receiving the voice output. The effect is that it becomes possible to do this. Further, the present invention is suitable for automatic vending machines, voice guidance devices, etc., which output a large amount of voice information in series from one voice output device, and which require prompt voice notification of changes in external or internal circumstances. It is.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an example of a vending machine to which the present invention is applied, FIG. 2 is a block diagram showing an example of the speech synthesizer shown in FIG. 1, and FIG. FIGS. 4A, B, and C are explanatory diagrams showing the audio output state of FIG. 1, respectively. FIG. 5 is a flowchart showing an example of the audio output control procedure of FIG. 1, and FIG. Audio data ROM in Figure 1
7 is a block diagram showing an example of the message data portion of the voice utterance table of FIG. 6. FIG. 1... coin mechanism unit, 2... sales control circuit, 3... product selection button, 4... product delivery mechanism, 5... voice generation control device, 6... voice synthesizer, 7... voice data ROM, 8...Amplifier,
9...Speech synthesis unit, 10...Speaker, 21-3
0...Control procedure, 41...Audio utterance table, 4
2...Table number, 43...Message data, 43a to 43d...Voice data, 44...Speech end END code, 45...Selection information, A,
B, C, D, E, F, G...Word data for each clause, _F1 ...Interruption flag, _F2 ...Speed up flag, _X1 to _X3 ...Word spacing adjustment timer.

Claims (1)

[Scope of Claims] 1 a A plurality of words are stored in advance for each clause or a predetermined number of words independently, and it is indicated whether or not audio output can be interrupted for each clause or for each predetermined number of words. a storage means that stores an interruption flag and a speed-up flag indicating whether speed-up of audio output is possible; b. selecting and specifying a necessary word from the plurality of words and requesting audio output of the word; a voice output requesting means; c. a selection means for selecting and reading out a corresponding word from the storage means in response to a request of the voice output requesting means; and d. converting the word read out by the selection means into voice. e. When receiving a new audio output request from the audio output requesting means while the audio outputting means is outputting audio, the interruptability flag and the speed of the words currently being outputted are set. and an audio output control means for suspending or speeding up the current audio output according to the up flag, and then permitting the new audio output. Generation control device.