JP2004334164A - System for learning pronunciation and identification of english phonemes "l" and "r" - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To solve the problems with the prior art that a learner finds it difficult to obtain a sufficiently satisfactory result when learning the pronunciation and identification of English phonemes "l" and "r" and that there is no other choice for the learner but to make such learning only on the limited learning site even if the learner desires to make study. <P>SOLUTION: The system is previously registered therein with the characteristic patterns of the voices "la" and "ra" uttered for several seconds by stressing the portions of the English phoneme "l" and "r" with lapse of time as model patterns and has a function to continuously display the model patterns and the imitation patterns of the learner in real time in compliance with the frequency of the learner in parallel on the same screen and to reproduce the voices of both as required. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  An information and communication network system that connects computers at language education sites, such as general English conversation schools and computers in junior high schools, high schools, universities, etc. that teach English at an early stage, and even individual learners in remote locations. By using this device, the present device can be used as a training device for pronunciation and identification.

  In teaching the pronunciation of the English phonemes "l" and "r", the conventional technology uses a mechanical method using hands, mouths, objects, drawings, etc., and an example of a native language speaker. We have adopted the method of showing as sound. Learners have learned by imitating. However, such mechanical and auditory guidance alone has not provided learners with satisfactory consent. On the other hand, little instruction was given to improve the discrimination ability of both phonemes. Further, since the area was technically unexplored, the network system was not used for the learning.

  In the prior art, in the training of pronunciation and discrimination of phonemes “l” and “r”, it has been difficult for the learner to give the learner sufficient learning results to the learner. There are three possible causes.

  First, it is very difficult for learners who only know Japanese to learn the pronunciation of both phonemes. Originally, both phonemes do not exist in Japanese. For this reason, Japanese have little or very little ability to discriminate between phonemes aurally. Therefore, even if you learn the pronunciation of both phonemes while listening to the instructor's model pronunciation, it is impossible for the learner to achieve a sufficient effect to be confident. This is similar to the difficulty that deaf people encounter when learning a language.

  Second, the prior art places little emphasis on biphone discrimination. Therefore, the training for discriminating the phoneme was not performed exclusively at the learning site. Thus, despite the need for accurate identification, learners could not learn it.

  Third, there is not enough space to obtain knowledge, techniques, and information for learning and acquiring the pronunciation and discrimination ability of both phonemes. In particular, it has been almost impossible for remote learners to achieve this goal. The following solutions are conceivable for these problems.

  For the first cause, this problem can be solved by visually supplementing the deficiency of the auditory function. In this apparatus, phonetic feature patterns (for example, waveforms, frequencies, amplitudes, etc.) of both phonemes, which cannot be obtained by the conventional technique, are visually displayed as a model pattern on a part of the computer screen. The learner simulates and inputs the sound while imitating while listening to the model sound. The input voice is analyzed in real time, and instantly arranged and displayed as imitation patterns on the rest of the same screen. Therefore, since the learner can easily compare and evaluate both patterns visually, accurate feedback can be obtained. Based on the obtained information, even if auditory information is scarce, sufficient pronunciation training can be performed.

  The second cause can be resolved in two stages. In the first stage, the learner accurately grasps the characteristic difference between the two phonemes from a perceptual and analytical viewpoint. In the second stage, training materials are compiled based on the identification method of a person whose native language is English (hereinafter referred to as “native person”). By using this material and training on the basis of the knowledge obtained in the previous stage, the learner can acquire not only the discriminating ability of both phonemes but also the discriminating ability similar to that of native speakers.

  As a first step, the apparatus is required for a learner with little discrimination ability to perceptually learn and acquire the features and differences of the two phonemes “l” and “r” using the processed voice. A device that provides assistance. Usually, we distinguish and utter various phonemes within a short time of 100 ms (millisecond or 1/1000 second), sometimes 20 ms or 30 ms, and instantly discriminate and perceive them. However, it is very difficult for beginners to acquire and discriminate between two phonemes that they have no experience with. However, by making each phoneme temporally stretched to produce a processed voice, the learner can easily acquire the perceptual features and differences of both phonemes. However, just acquiring it is meaningless. In fact, it must be able to be fully utilized in the field. Therefore, the apparatus includes various processed voices, basic voices and words for discrimination training, and processed voices composed of both phonemes, basic voice and word pairs, and model patterns to facilitate comparison of both phonemes. In addition, in order to make it more practical, a large number of pronunciation examples by many general natives are registered. The apparatus has a function of repeatedly reproducing these sounds.

  As a second step, the present apparatus is a training apparatus for bringing the English learner's ability to discriminate phonemes "l" and "r" closer to that of a native language speaker. For this purpose, the apparatus includes a plurality of artificially synthesized voices in which English voices “la” and “ra” are artificially synthesized based on the identification standards of native speakers, and various types of voices that are performed using the synthesized voices. Identification training form is registered. In addition, the following various functions are provided for efficient training. A function for presenting each synthesized voice in various ways, a voice reproduction function for reproducing each synthesized voice, and a learner using a mouse or a keyboard to determine whether the presented voice was “la” or “ra”. A function of identifying and inputting information, a function of totalizing the input information and determining whether the information is correct, a function of reporting the result, and the like.

  For the third cause, the present apparatus provides similar support to a remote learner by using a network system, that is, communication means for connecting computers such as the Internet and a LAN. It has a function to provide fairly.

  This device combines the model patterns of the English phonemes “l” and “r”, which do not originally exist in Japanese, and the imitation patterns of both phonemes of the learner in one real-time and simultaneously It can be presented on the screen in parallel. Therefore, the learner could easily compare and contrast, and enabled self-evaluation. As a result, the learner can compensate for his own auditory deficits, can perform training independently, and can easily learn the pronunciation method of both phonemes.

  This apparatus adopted processed speech in discrimination training of phonemes “l” and “r”. As a result, it became possible to easily learn the characteristic differences between the two phonemes. In addition, the characteristics of short-term memory were utilized so that learning and memory could be performed efficiently, so that the timing of a basic voice pair and a word pair could be set and reproduced and presented. Furthermore, a plurality of artificially synthesized speeches “la” and “ra” organized based on the identification criteria of the native language were adopted as training speech materials. As a result, learners can now approach their own identification criteria to those of native speakers.

  In addition, the device does not only assist in acquiring the characteristic differences between the two phonemes as mere knowledge. In order to make use of this knowledge in actual situations, various training methods were devised, and various discriminative training tests were created. By performing these training tests, learners can smoothly acquire practical discriminating ability.

  Finally, this device uses a network system, which is a means of communication using a computer, so that not only a limited place such as a school in a broad sense but also a wide range of learners who want to learn The training of discrimination can be freely carried out only for the time corresponding to.

  In learning the pronunciation and discrimination ability of the phonemes “l” and “r”, the present invention can compensate for insufficient information using only auditory information as visual information. That is, analytic information (characteristic differences between the two detailed voices) mechanically obtained using the voice analyzer is visually presented as a pattern on a computer screen. Therefore, the learner can clarify and acquire the characteristics and differences between the two phonemes even if the discrimination ability by hearing is insufficient and the visual ability is sufficient. In addition, by conducting effective pronunciation and discrimination training using various functions and audio materials, the learner can increase the acquired differences to practical abilities rather than mere knowledge. In addition, the network system allows the learner to receive the learning while staying there.

  A brief description will be given of the figure adopted first. FIGS. 1, 2 and 3 are schematic diagrams of both voices created based on the voice analysis results of English voices “la” and “ra”. The horizontal axis represents time t (unit: 1/1000 second, hereinafter referred to as “ms”), and the vertical axis represents frequency (unit: Hz). In these figures, in order to make the difference between the two voices as clear as possible, parameters are created and schematized by focusing only on the characteristic portions of both voices. Therefore, the numerical values used are rough, and the actual speech is much more complex and varied.

  Both voices are composed of three parts along the time axis. A portion where only the phoneme “l” or “r” is pronounced during the first 0 ms to 100 ms, and a portion where the vowel “a” is added and “la” or “ra” is pronounced during the next 100 ms to 150 ms During the remaining 200 ms, only the vowel "a" is uttered for a relatively long period of time, consisting of three parts.

  Note that F1, F2, and F3 are called formants and represent some of the basic elements that make up human speech. Usually, there are four to five, and a frequency hertz (Hz) is used as a unit. Then, F1, F2, and F3 are described in order from the one with the lowest frequency. In describing the features of the sounds “la” and “ra”, only the above three elements were adopted, and the remaining elements were omitted because they were not important. FIG. 3 shows only F3, and F1 and F2 have been omitted for the sake of simplicity in view of showing similar temporal changes in both phonemes.

  FIG. 4 and FIG. 5 are implementation diagrams when English voices “la” and “ra” are actually displayed on a computer screen as model patterns. However, in each of the figures, the phoneme "l" or the phoneme "r" is uttered while being emphasized temporally more than actual. FIGS. 6 and 7 show reference patterns when both phonemes come in the middle of a word and when they come last. FIG. 6 shows a pattern of “alive-arrive”, and FIG. 7 shows a pattern of “hail-four”. FIG. 8 is a diagram illustrating an auditory change of both phonemes with respect to a time axis.

  In learning pronunciation, it is important for learners to know the pronunciation characteristics of native speakers. As seen in FIGS. 1 and 2, unlike English, English has a habit of producing only consonants as single tones for a relatively long time (approximately 80 ms to 150 ms). Therefore, it is important for a learner who wants to learn English to keep this point in mind so that both phonemes can be continuously pronounced as single sounds. An embodiment for obtaining pronunciation of both phonemes will be described below in the order of "l" and "r".

  The method invented this time makes it possible to visually grasp the result even without the auditory discrimination ability by visually displaying the analysis result of the voice. As shown in FIG. 4, the phonetic "l" is continuously pronounced for about 2 to 3 seconds by the native language person, and the result is analyzed and displayed as a model pattern on a part of the computer screen. On the other hand, the learner pronounces and inputs the model voice reproduced at the same time through the microphone to the computer based on the instructor, the instruction book, or his or her already learned pronunciation knowledge. The input phonemes are instantaneously analyzed and displayed on the remaining portion on the screen (the lower half on the screen in FIG. 4) as the imitation pattern in parallel with the model pattern already displayed. The learner can easily compare visually (and, if necessary, auditory) by comparing his or her imitation pattern with the model pattern. Moreover, even if the two phonemes cannot be distinguished by hearing, it is possible to objectively verify whether one's pronunciation is good or not if there is a visual discrimination ability.

  An important point to note here is that the difference between the frequencies of F2 and F3 (hereinafter referred to as "gap") during the first 100 ms when the phoneme "l" is pronounced, as seen in FIG. Is constant and hardly changes. Although there are individual differences, the gap between both formants is usually on average about 1300 Hz to 1900 Hz for native speakers. That is, when the native speaker emphasizes and pronounces a word including “l”, the above gap is maintained for about 80 ms to 150 ms. Therefore, it is preferable that the learner also exercises pronunciation so that a gap close to this can be maintained for a certain period of time.

  Further, according to the present invention, a learner can utter a voice in real time to analyze the voice and display the voice as an imitation pattern. Therefore, the learner can compare the two patterns instantly, find a problem, and continue the training smoothly at his own pace through trial and error. Effective training to achieve the desired gap and duration can be conducted with the assistance of leaders and the like as needed. Ultimately, the learners themselves can check their training results with their own eyes, so they can achieve satisfactory results.

  When the phoneme “l” can be pronounced, the program moves on to the practice of the voice “la”. The positional relationship between the tongue and the mouth is moved from the state of "1" to the positional relationship of the vowel, and at the same time the vowel is pronounced. For example, if a vowel "a" is pronounced, an English voice "la" is obtained. When we actually analyze the voice "la" in time, we can see the process of transitioning first to the phoneme "l", then to "la", and finally to "a", as seen in FIG. I can do it. While watching such a model pattern, practice repeatedly to obtain a similar pattern. In order to stably obtain a favorable gap, it is important to continuously train while comparing and imitating the model and the imitation. Note that the gap between F2 and F3 slightly changes depending on the type of the vowel that comes next. However, it is preferable to keep it broad as in the case of "a".

  Next, the description moves to the phoneme “r”. FIG. 2 shows that only the phoneme “r” is generated during the time from 0 ms to 100 ms. “R” is a major feature in that the gap between F2 and F3 is much narrower than “l”. However, the frequency of F2 has no remarkable difference between the two phonemes, and is about 900 Hz to 1200 Hz on average. Therefore, when the phoneme "r" is pronounced, the frequency of F3 (note, not a gap) Is much lower than in the case of “1”, and it is important to average from 1500 Hz to 1800 Hz. Some native speakers have a frequency exceeding 2000 Hz as a phenomenon particularly observed in women. However, in the case of men, it is generally between 1500 Hz and 1600 Hz, and cases exceeding 1800 Hz are rare.

  FIG. 5 shows, on the computer screen, a model pattern of a sound “ra” in which the phoneme “r” is particularly sustained for 2 to 3 seconds, as in the case of the phoneme “l”. It is. The learner pronounces while imitating the model sound played simultaneously based on the instructor, the instruction book, or his already learned pronunciation knowledge. The learner trains according to the same process as in the case of “l” while illuminating his or her imitation pattern displayed on the rest of the screen with the model pattern. Once the "r" can be pronounced, it begins to be more practical training with vowels. That is, if the positional relationship between the tongue and the mouth is shifted from the position of “r” to the position of a vowel, for example, the positional relationship of “a”, and the vowel “a” is pronounced at the same time, the English voice “ra” can be obtained. .

  The present invention enables independent evaluation of the pronunciation results of one's own pronunciation only with the visual discrimination ability without the auditory discrimination ability of both phonemes. In the prior art, learners have had to rely on the evaluation of others (eg, instructors). For this reason, even if it is judged that the pronunciation ability acquired is perfect, it is difficult to have 100% confidence because there is anxiety about "Is this really good?" Met. The reason is that she lost her auditory discrimination ability between the two phonemes soon after birth because she grew up in a Japanese environment. However, the present invention makes it possible for a learner to easily compare his or her own pronunciation results on a screen with a model at any time, even if he has no discriminating ability. Therefore, the learner can make an objective evaluation himself. By taking necessary measures based on the obtained evaluations and incorporating them into the next and subsequent drills, it was possible to smoothly achieve results.

  Training can be performed in the same manner as in the case of basic speech, regardless of whether both phonemes are at the beginning, middle, or end of a word. The preceding discussion has discussed the basic speech consisting of both phonemes. However, when both phonemes come to the beginning of a word (for example, “rice” or “rice”), even when they come at the middle and at the end, a pattern similar to that shown above is formed. The first case is exactly the same as the case of the basic sound, so the explanation is omitted. FIG. 6 shows the patterns of the words “alive” and “arrive” in the middle, and FIG. 7 shows the patterns of “hal” and “four” in the end. In order to facilitate comparison of both phonemes, the case where the upper part is “l” and the lower part is “r” is shown. As can be seen from the figures, the portion surrounded by two vertical lines in each figure is the phoneme “l” or “r”. That is, it can be seen that the characteristics of both when the two phonemes come in the middle of the word have exactly the same characteristics as those when they come at the head. In the last case, the frequencies of F2 and F3 are slightly higher for the phoneme "r", but the gap between them is almost the same as in the first and middle cases. Therefore, it is important that the learner trains with this point in mind. It should be noted that the present apparatus registers a large number of words in which both phonemes are at the first, middle, or last, and a large number of pairs of words that facilitate comparison of both phonemes.

  Next, a description will be given of the discrimination training device. The difference between the two phonemes “l” and “r” can be captured from two points, a perceptual point of view by a sensory organ and a voice analysis point of view by a measuring machine. Therefore, the present invention is configured in three stages: learning the basics of discriminating both phonemes from each viewpoint, and further implementing training. The first stage learns where the characteristic differences between the phonemes "l" and "r" are perceptually. The second step is to learn how to identify native speakers using artificially synthesized sounds created based on analytical differences. Further learning the differences does not help in the actual situation. Therefore, as a third step, a large number of training tests created using various audio materials are performed to acquire practical skills. The embodiment will be described in the above three steps in this order.

  In the first stage, two methods were employed. One is a processed voice obtained by temporally stretching a basic voice (refers to a voice composed of “consonants + vowels” such as “le”, “ro”, “ble”, and “blo”). Used. The other uses a pair of phonemes composed of basic sounds (for example, lu-ru, ple-pre, etc.) and a pair composed of words (for example, play-play, etc.). However, the two sounds constituting each pair are arranged so as to be reproduced at a specific time interval.

  First, the use of processed speech has made it possible to clarify the characteristic differences between the two phonemes. Usually, we distinguish and pronounce various sounds in a very short time, such as 20 ms or 30 ms, with exceptions. However, it is almost impossible for an inexperienced learner such as a Japanese to accurately grasp the difference between the two phonemes “l” and “r” generated instantaneously. One of the optimal methods for solving this problem is to use both basic sounds, for example, processed sounds obtained by temporally extending “la” and “ra”. By this operation, the process in which the two voices change along the time axis becomes slower than the actual time, and the voices can be presented in a form in which the difference can be easily perceived. That is, instantaneous differences between the two phonemes that are normally missed or hard to hear are clearly displayed. Therefore, the learner can easily learn the characteristic difference between the two phonemes.

  In fact, when both voices are heard, the phoneme “l” changes rapidly, but the phoneme “r” changes slowly. FIG. 8 is a diagram illustrating this difference. The horizontal axis represents time (unit: ms), and the vertical axis represents the process of changing both phonemes. Also, between the two levels on the vertical axis (one phoneme "l" or "r" and another vowel "a"), each phoneme starts with its own phoneme, and the speech "la" or "ra" ”To the vowel“ a ”. The solid line indicates the changing process of the phoneme “l”, and the dotted line indicates the changing process of the phoneme “r”. As can be seen from this figure, the characteristic difference between the two phonemes is in the transition process from each phoneme to the vowel "a". In the case of the phoneme "l", the sound rapidly approaches the vowel at first, but thereafter changes asymptotically and finally reaches the vowel. On the other hand, in the case of the phoneme “r”, it changes gradually at first, changes rapidly in the latter half, and finally reaches a vowel. It is difficult to accurately and analytically compare such a symmetrical difference between the two because of the accuracy of the measuring instrument and the individual difference. However, as a rough numerical value, the phoneme “l” is better than the phoneme. It almost approaches a vowel in a time that is about one-third to one-third shorter than “r”. In the case of FIG. 8, the vowel "a" is approximated by multiplying the phoneme "l" by 15 ms from 100 ms to 115 ms, while the phoneme "r" is multiplied by 40 ms from 100 ms to 140 ms.

  The other is to perform efficient learning by using training materials based on human memory skills. Using two words composed of both phonemes (for example, frame, frame, etc.), a pair of audio materials is created so that both words can be reproduced and presented at an interval of 1/2 second to 1 second. Further, the timing is set so that this pair can be reproduced and presented twice within 7 to 8 seconds. On the other hand, in the case of basic speech, a pair is created in the same manner as in the case of a word. However, when reproducing and presenting, the timing is set to such a speed that the created pair can be repeatedly reproduced twice or three times in 7 to 8 seconds. The pairs of words and basic speech set as described above are continuously and repeatedly presented as training materials. The difference between the two phonemes learned by the processed speech is ensured by this method, and can be effectively stored based on the characteristics of short-term memory.

  In the second stage, native speakers make a special identification in terms of frequency. With emphasis on this point, it is important to bring the style of recognition closer to that of native speakers in order to acquire practical skills. As has been clarified in the above description, an important difference between the two phonemes is remarkably expressed as a difference in change of F3 with respect to the time axis (between 0 ms and 150 ms in FIGS. 1 and 2). . When F1 and F2 are actually kept constant and the vibration value of the start of F3 (hereinafter referred to as “initial value”) is variously changed experimentally, the native language person assumes that the initial value has a certain frequency ( On the basis of the average (between 2200 Hz and 2400 Hz), a lower case is identified as "r", and a higher case is identified as "l". Therefore, when it is desired to learn an identification style similar to a native language person, it is preferable to acquire such an identification standard for a native language person.

  Focusing on this point, in the present invention, the parameters of F3 as shown in FIG. 3 are set using artificially synthesized speech whose initial value is easily changed. The initial value of F3 was divided into five equal parts at intervals of 250 Hz from a minimum value of 1600 Hz to a maximum value of 2850 Hz, and a total of six pieces (1600 Hz, 1850 Hz, 2100 Hz, 2350 Hz, 2600 Hz and 2850 Hz) were set. Then, a temporal change is added as in FIGS. That is, each initial value is maintained from 0 ms to 100 m, linearly changes from the individual initial value to 2400 Hz from 100 ms to 150 ms, and 2400 Hz is maintained from 150 ms to 350 ms. FIG. 3 shows the parameters organized in this manner. That is, it shows that six parameters of only F3 are collected and collected in one figure. To each of these different F3s, the common F1 and F2 parameters for both phonemes were added to synthesize a total of six types of artificial speech. In addition, in order to create a synthesized voice closer to the real voice, it is also possible to perform synthesis by taking into account the formant parameter value and the number of synthesized sounds.

  In order to use these synthesized speeches as discriminative training materials, it is necessary to classify each speech as "la" or "ra" again by the native language. In classification, it is preferable to set an initial value so that the number of both voices is the same. The synthesized speeches “la” and “ra” thus classified and used are used as data for actual discrimination training. In this description, for convenience, each synthesized voice is shown in FIG. And the obtained synthesized speech is used as data for discrimination training.

  In the third stage, it is necessary to conduct training to increase the knowledge acquired in the first stage to the ability to be used in reality. In everyday life, we understand how many people can speak in each sound quality, and what they say instantaneously. It is important to construct and implement a training method paying attention to this point. Therefore, in the present invention, in both the first and second stages, a number of pronunciation examples of a large number of native speakers (pronunciation examples of less than 50 native speakers are registered in this device). The training materials, that is, processed voices, basic voices, words, artificially synthesized voices, etc., are registered in the present apparatus as voice materials for training tests in the form of single sounds or in pairs. Note that the adopted word group may be knitted in a manner such as "late-rate" or "glass-glass" so as to always form a pair with the corresponding word for the phonemes "l" and "r". It is selected and registered so that it can be done. In addition to this, a single sentence or a compound sentence containing one or more words consisting of “l” or “r” in the sentence including the single sentence is also registered as the material.

  The form of the training test to be performed varies depending on the type and form of the sound to be reproduced and presented. Therefore, in the present invention, each test method is devised for a case where a reproduced sound is a single sound, a pair, and a sentence.

  In the case of a single sound, one sound is reproduced between 1400 ms and 1500 ms as a presentation method. The learner clicks the answer example on the screen to answer whether the sound reproduced using the mouse contains the phoneme “l” or “r”. At the same time as the answer, the reaction time required for the answer is measured in milliseconds. At the same time as answering, a correct signal or an incorrect answer is signaled by sound or an illustration is fed back for 1000 ms. If you want to hear it again before answering it, you can click the "Replay" button to play it again and listen. Do not move to the next trial unless you answer. When the feedback ends, the process moves to the next trial at the same time.

  In the case of a pair, one pair is reproduced and presented in 2800 ms to 3000 ms. The answering method differs depending on the relationship between the two voices forming the pair. For example, in the case of a pair in which the remaining parts except for the phonemes “l” or “r” are pronounced exactly the same (for example, “lock-rock” or “rice-rice”, etc.), the learner presents using the mouse. It is determined whether the pair formed is composed of the "same" phoneme or "difference", and an answer is made by clicking the corresponding column on the screen. Alternatively, when the type of the pair is other than the above (for example, “lack-rick” or “rack-lick”), the order of both phonemes included in the voice adopted for each pair is “lr”, “r−r”. 1 "," rr ", or" ll "is determined, and the corresponding screen is clicked to answer. Alternatively, by paying attention to the phoneme "l" of the two voices forming a pair, it is determined whether the voice including the phoneme is reproduced "before" or "after", and the answer is similarly given. The transition of the reaction time, replay, feedback, and trial is performed in the same manner as in the case of the single tone.

  If the presented voice is a sentence, if it contains only one phoneme, it is determined whether it is "l" or "r", and click on the phoneme name on the screen as above. Answer. If a plurality of the phonemes are included, the phonemes of the phoneme are sequentially ordered from the front, such as "l, l, ...", "r, l, ...", "r, l, l, ...". Select the order on the screen and click to answer or press the key of the phoneme name from the keyboard to answer. The transition of the reaction time, replay, feedback, and trial is performed in the same manner as in the case of the single tone.

  The number of sound types used in the test is about 20 to 30 in the case of a single sound in one test, and these sound types are repeatedly reproduced four to six times in one test. Therefore, in one test, the reproduction of the material sound is tried about 100 to 130 times. On the other hand, in the case of a pair, about 15 to 30 types of audio materials are repeatedly reproduced twice or five times. The texts are organized so that one test can be completed in around 10 minutes, depending on the length of the text. In any case, the audio reproduction order is random, and the audio data is set so as not to be reproduced in the same order twice even if the same test is continuously performed.

  You can create countless types of tests by changing the type of sound and the number of trials. Therefore, the learner can select a favorite test and conduct training. For example, consider only the case where a word is reproduced and presented as a single sound. One test has the same number of voices composed of “l” and the same number of voices composed of “r”, and the sound types are configured to be paired with each other. Therefore, when creating a test composed of 20 kinds of sound types, there are a total of 120 or more words in which “l” and “r” form a pair in the whole English word. There are countless combinations of only 20 combinations taken out at random. Further, it is possible to construct various training test groups by adding basic speech, pairs, synthesized speech, sentences, and the like. Therefore, the learner can select and extract his or her favorite test from a myriad of test groups (actually, about 200 types are registered in the present apparatus) and execute it for a desired time without getting tired. Increase training efficiency.

  In the discrimination training test, it is essential to shorten the time required for answering, not to mention the correct answer rate. Learners must be able to accurately identify the differences between the two phonemes. But it can be impractical if it takes seconds to make a decision. Therefore, the answer time or the reaction time is significant in each discrimination test. At the same time as raising the correct answer rate to about 100%, the learner is required to reduce the reaction time from 1000 ms to 1300 ms as a guide in the present apparatus. Conduct discrimination training so that the above standard can always be achieved on average. The apparatus has functions for performing the above-mentioned training, for example, a function for setting an answer screen, a function for reproducing and presenting voice, an answer input function, a reaction time measuring function, a replay function, a feedback function, and the like. Have.

  In addition, the present apparatus can implement and provide similar training to learners at geographically distant places by using a network system performed using a computer. Files required to start and execute the device on the server or host computer, such as "sound files" as materials, "test execution files" in which various training test execution procedures are registered, and application files for executing tests Is mounted in advance. Each learner at a remote location connects his computer to the server via the Internet or LAN. The learner installs the application on his / her computer in advance, or downloads and starts the application file. By operating according to the instructions while viewing the screen of the opened application, it is possible to automatically download a desired test execution file from the server and execute the test. Each test result is automatically saved on the server as a file. If necessary, the learner can browse his / her own results from the application. It should be noted that the application file is set so as to automatically read the file on the server when opened. In this way, it is possible to carry out the same training as when not using a network. Although a slight time delay is expected depending on the communication speed, it is considered that this disadvantage can be solved by future technological innovation.

  If the analysis accuracy of speech, the synthesis accuracy, and the display technology are improved with the advance of technology in the future, the pronunciation and discrimination training method will be improved accordingly, and it will be possible to carry out the training more efficiently. The present invention is fully capable of incorporating the results of those anticipated technological advances as part of the training material without changing the basic concept of the device.

  The basic concept adopted in the present invention is not limited to the phonemes “l” and “r”. Generally, when learning a foreign language, the concept can be applied to learning a new voice that is not in the native language. That is, the learner easily obtains the new speech analytic and perceptual features by converting them into a perceptible presentation method as necessary, as in the case of “l” and “r”. Can do things. Based on the obtained basic knowledge, the pronunciation and discrimination training of the phoneme can be performed by a method similar to the training method adopted in the present device.

Schematic diagram of English voice "la" Schematic diagram of English voice "ra" Schematic diagram of F3 parameters of artificially synthesized sounds Model pattern of phoneme "la" Model pattern of phoneme "ra" Model pattern when phoneme "l" or "r" comes in the middle of a word Example pattern when phoneme "l" or "r" comes to the end of word The process by which the phoneme "l" or "r" shifts to the vowel "a"

Explanation of reference numerals

Hz Frequency t Time (1/1000 second)
F1 Formant 1
F2 Formant 2
F3 Formant 3

Claims (6)

The present apparatus is designed so that a learner trying to learn the pronunciation of English phonemes “l” and “r” can objectively convert the model pattern of a person whose native language is English and the model pattern of the learner, including voice. This is a device that enables the user to simultaneously practice pronunciation while comparing and contrasting with the above. For this purpose, the present device pre-emphasizes the characteristic patterns of the sounds “la” and “ra” uttered for a few seconds by temporally emphasizing the parts of the English phonemes “l” and “r” as model patterns. It is registered. In addition, a function for displaying the model pattern and the imitation pattern of the learner in parallel on the same screen in real time and continuously in accordance with the frequency of the learner, and reproducing both sounds as necessary. It has functions for: This device is an identification assisting device that emphasizes the characteristic differences between the English phonemes “l” and “r”, expresses them more clearly, and allows the learner to quickly learn the features of both phonemes. . For this purpose, the present apparatus pre-processes a basic voice (for example, “li” or “re”) of a person whose native language is English, including phonemes “l” or “r”, and preprocesses the processed voice. It is registered. Furthermore, in order for the learner to compare and contrast the two phonemes, grasp the difference between them, and to facilitate the storage, the processed voice, the basic voice, and the word may be used as a single sound or as a pair (for example, li-ri, bright-bright). Etc.) Many audio materials are registered so that they can be reproduced at specific time intervals. In order to acquire more practical discriminating ability, a sentence containing either or both of the phonemes is also registered as the presentation voice. The present device is also a training device for making the discriminating ability of phonemes “l” and “r” of an English learner close to the discriminating ability of native speakers. For this purpose, the apparatus previously generates a plurality of artificially synthesized voices based on the voice analysis results of the English voices “la” and “ra”, and re-generates them based on the native language identification criteria. Audio materials classified as “la” or “ra” are registered. The present apparatus is also a training apparatus for efficiently operating various registered materials described in claims 2 and 3. Therefore, it is designed to maximize the efficiency of organizing the materials, presenting the materials, the learner's answer method, the type and presentation of the feedback, and reporting the results according to the purpose. This apparatus is intended to enhance the ability to identify English by using a computer-based network system (for example, a communication means that connects computers over a wide area via the Internet or a limited area via a LAN). It provides the same opportunities and support as remote training for remote learners. The speech training method adopted in the present device can also be applied to a case where a learner of a foreign language other than English learns and acquires a new speech that is not in his / her own language. The basic concept of the present invention, i.e., new speech features, is analyzed perceptually and analytically and provided to learners in an auditory and visually understandable manner. Based on the acquired basic knowledge, the learner performs training of pronunciation and identification of the voice in the same manner as the method adopted in the present invention. Without changing the basic concept, it is possible to easily achieve the purpose of acquiring a new voice using the present device. It should be noted that the present apparatus can incorporate the analysis results of the new technology without changing the basic concept of the present invention with respect to the expected advance of the voice analysis technology.