JPH0731472B2 - Speech language practice machine - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (A) Technical Field of the Invention The present invention utilizes a phonetic language training machine, and in particular, continuous word voice recognition, so that it is possible to effectively carry out foreign language conversation training and grammar training. It's about training machines.

(B) Conventional technology and problems In language education, especially in learning foreign languages, it is important not only to read, write, and listen, but to actually speak and speak. Conventionally, as a speech language practice machine, for example, one using a tape recorder has been used, but according to the conventional speech language practice machine, the speech reproduced from the tape teaching material or the like is simply repeated or reproduced. The learner uttered a part of the spoken text in other words. However, since the learner himself / herself has to judge the correctness of the voice uttered by the learner, it is often the case that the learner does not notice it even when the learner utters it by mistake. Moreover, it was impossible to give a proper guideline to the learner's error.

(C) Object and Structure of the Invention The present invention aims to solve the above problems, and considers that the language can be effectively learned by uttering the entire sentence. Automatically judge whether there is any, and if there is an error, point it out and provide a means to give appropriate guidelines, especially to be able to effectively learn conversation and grammar of foreign languages etc. The purpose is to Therefore, in recent years, we have focused on applying the technology of speech recognition processing, which is rapidly advancing in recent years, especially continuous word speech recognition, and aim to improve the language education effect. That is, the phonetic language practice machine of the present invention includes a lesson storing means for storing at least the content of the lesson including information about question content and answer content, and an instruction means for designating a word used in the lesson stored by the lesson storage means. And a registration means for registering the voice of the word designated by the above-mentioned instruction means uttered by the user, an output means for presenting the question content by voice synthesis or the display of characters / figures according to the content of the instruction, and the output The continuous word voice recognition is performed based on the voice feature parameters of the registered words that are registered in advance and are allowed to be input, by inputting the voice word string that is continuously uttered for the question content presented by the means. With respect to the speech recognition means for performing the speech recognition and the recognition result by the continuous word speech recognition, the information on whether the answer is correct or incorrect by referring to the course memory means or the information on the error portion It is characterized in that a guidance output unit for outputting. Embodiments will be described below with reference to the drawings.

(D) Embodiment of the invention FIG. 1 is a block diagram of one embodiment of the present invention, and FIGS. 2 and 3 are explanatory views of processing modes of one embodiment of the present invention, respectively.
FIG. 5 shows an example of a state transition diagram of a finite state automaton, and FIG. 5 shows an example of a storage format of the syntax information storage unit shown in FIG.

In the figure, 1 is a main control unit, 2 is a microphone, 3 is a parameter extraction unit, 4 is a parameter registration unit, 5 is a word dictionary unit, 6 is a continuous word speech recognition unit, 7 is a syntax information storage unit, and 8 is an output unit. Reference numeral 9 is a voice synthesis unit, 10 is a display control unit, 11 is an amplifier, 12 is a speaker, 13 is a display, 14 is a teaching memory unit, and 15 is a word designating unit.

The main control unit 1 executes an instruction given in advance by a processor and drives various processing units described below. Voice input is performed from the microphone 2. The parameter extraction unit 3 acoustically analyzes the voice signal input from the microphone 2 and extracts an input feature parameter time series representing the feature of the input voice. For example, frequency analysis is performed using a band filter group or the like to extract the moment M ₁ corresponding to the first formant frequency, the moment M ₂ corresponding to the second formant frequency, and further the low range power and the high range power. , The sample points regarding these feature quantities are determined, and the time series information of the feature quantities is obtained.

The parameter registration unit 4 registers, in the word dictionary unit 5, the characteristic parameters extracted from the learner's utterance for a word used for learning that is designated in advance for voice recognition performed later, together with the item name of the word. Is. The continuous word speech recognition unit 6 analyzes the feature parameters of the learner's response analyzed and extracted by the parameter extraction unit 3 in units of words stored in the word dictionary unit 5 and the so-called two-stage DP matching or It recognizes continuous word speech by matching using the level building (LB) method. When recognizing the continuous word voice, the storage information of the syntax information storage unit 7 is used as necessary. The syntactic information storage unit 7 is preliminarily prepared with syntactic information such as the permitted degree of continuation of words according to the content of the course. By using this syntactic information at the time of recognition, the recognition speed and the recognition rate are improved. There is.

The output unit 8 presents the content of the guide, the content of the question, the content of the answer, etc. to the learner, and includes, for example, a voice synthesis unit 9 and a display control unit 10. The voice synthesized by the voice synthesizer 9 is amplified by the amplifier 11 and output from the speaker 12. The display control unit 10 uses a display such as a CRT or a liquid crystal display.
It controls the display to 13.

The contents of the lesson including guide contents, question contents, and answer contents are read from the lesson memory unit 14 under the control of the main control unit 1. The content stored in the course memory 14 is, for example,
It can be replaced by tape, flexible disk, etc. The word instruction unit 15 extracts words used in the lesson from the content of the lesson storage unit 14 and gives an instruction via the output unit 8.

Next, the processing mode of the embodiment shown in FIG. 1 will be described.
FIG. 2 shows an example of a learning method that does not use syntax information.

First, at the start of learning, a process 20 is used to instruct a word to be used in the lesson and prompt the learner to speak. When the learner utters the instructed word, the characteristic parameter is extracted from the voice and the word is registered by the process 21. Next, the contents of the question are displayed by processing 22. For this question,
When the learner responds with a voice, the voice is input by processing 23 and a series of characteristic parameters are extracted. This feature parameter and the feature parameter in the word dictionary registered in the above process 21 are used in process 24 to perform continuous word speech recognition using, for example, two-stage DP matching. In the two-stage DP matching, as is well known, the collation process is decomposed into a word unit process and a whole process, and each of them is efficiently executed by dynamic programming.

When the recognition result is obtained, the recognition result is displayed by processing 25. Then, in the process 26, information indicating whether the answer is correct or incorrect and pointing out an error portion are notified by voice, for example, by voice synthesis. Next, in process 27, it is determined whether or not there are still more questions to be answered, and if all the questions are completed or if there is a stop instruction from the learner, the process is completed. If the question is still left and the learner wants to continue, the control is returned to the process 22, and the next question is repeated in the same manner.

FIG. 3 shows a processing mode of an embodiment in which syntax information is used for continuous word speech recognition. FIG. 3 process 30 to 33
Is the same as the processing of processing 20 to processing 23 shown in FIG. However, in the question of the process 32, the question of replacing words in the sentence is given. Generally, in the case of such a problem, the continuation relation of the preceding and following words is predetermined. That is, only the arrangement of words according to a certain syntax is allowed, and arbitrary words are not used separately. In the speech recognition in process 34, such syntax information is used. Specifically, the following information is used, for example.

For example, when the sentence pattern “I go to church.” Is given, the permissible state transition in the form of replacing this subject with another pronoun is as shown in FIG. Note that such a state transition diagram is known in a finite state automaton. This information is stored, for example, as a state transition table of syntax information as shown in FIG. The contents are as follows. In the start state (0), "I", "Yo
Input of "u", "They", "He", and "She" is allowed. Then, when any one of "I", "You" or "They" is input, the state (0) transits to the state (1). On the other hand, if “He” or “She” is entered, the status (2)
Transition to. In state (1), input of "go" is permitted, and in state (2), input of "goes" is permitted. The same applies hereinafter, and the state (6) is the final state.

The continuous speech recognition using the syntactical information as shown in FIG. 5 is performed, for example, in October 1980, the Acoustical Society of Japan, Proceedings of the Acoustical Society of Japan "1-1-23 Continuous speech recognition system by the two-stage DP method with automaton control. , Etc. For example:

Input pattern A, standard pattern B ⁿ And Here, the partial pattern A (l,
m), Is extracted and the distance between the partial pattern of the input pattern A and the standard pattern B ⁿ is obtained.

D (l, m, n) = D (A (l, m), ^Bn ) The syntax information of the automaton is determined by α = <K, Σ, Δ, P ₀ , F>. Here, K is a set of states, Σ is a word set, a set of “I, You, ..., to, church”, Δ is a state transition table as shown in FIG. 5, p, q Is expressed as q = δ (p, n) (where n is a word number), P ₀ is the initial state (= 0), and F is the final state (= 6).

In the second stage process of the second stage DP, the initial condition is T (0,0) = 0 T (m, q) = ∞ (when m ≠ 0, q ≠ 0), and the following recurrence formula is solve.

However, q = δ (p, n), and at the same time the optimal value of l, p, n Respectively, Save as. The final state is used as the judgment process. (However, Means to calculate a variable q that gives the minimum in []. ) age, Then, As a result, the following equation is repeated. By the above processing, the word string as the recognition result is obtained in reverse order.

When the learner's voice is recognized by the processing 34 shown in FIG. 3, the recognition result is displayed by the processing 35. And processing
With 36, a pointer is given, for example, by voice synthesis. Then, according to the determination processing 37, the processing 32 is performed until the question is finished.
Or, the process 36 is similarly repeated. By using the syntactic information, more accurate recognition can be performed quickly. Of course, the automaton-controlled two-stage DP matching is one embodiment, and other continuous word voice recognition methods may be used.

(E) Effects of the Invention As described above, according to the present invention, by applying the continuous word speech recognition technique, the learner's utterance for the entire sentence is input, the correctness is automatically determined, and conversation practice of a foreign language is performed. ， Become able to effectively practice grammar. In particular,
By instructing the learner to utter a word in advance, the learner can be made aware of the words related to the course, and it is possible to recognize a specific speaker and a specific word. Good results in recognition rate and recognition speed. Is obtained.

[Brief description of drawings]

FIG. 1 is a block diagram of one embodiment of the present invention, and FIGS. 2 and 3 are explanatory views of processing modes of one embodiment of the present invention, respectively.
FIG. 5 shows an example of a state transition diagram of a finite state automaton, and FIG. 5 shows an example of a storage format of the syntax information storage unit shown in FIG. In the figure, 1 is a main control unit, 2 is a microphone, 3 is a parameter extraction unit, 4 is a parameter registration unit, 5 is a word dictionary unit, 6 is a continuous word speech recognition unit, 7 is a syntax information storage unit, and 8 is an output unit. Reference numeral 9 is a voice synthesis unit, 10 is a display control unit, 11 is an amplifier, 12 is a speaker, 13 is a display, 14 is a teaching memory unit, and 15 is a word designating unit.

Claims (2)

[Claims] 1. A course storage means for storing the content of a course including at least information about question content and answer content, an instruction means for designating a word used in the course stored in the course storage means, and a user's utterance. The registration means for registering the voice of the designated word by the above-mentioned instruction means, the output means for presenting the question content by voice synthesis or the display of characters / figures according to the instruction content, and the output means for presenting A speech recognition means for inputting a continuously uttered speech word string in response to a question content and performing continuous word speech recognition based on speech feature parameters of registered words registered in advance and allowed to be input at this time. As for the recognition result by continuous word speech recognition, a guideline output hand that outputs information about whether the answer is correct or incorrect or information about an error point by referring to the above-mentioned course memory means. Voice language trainer aircraft, characterized in that it was equipped with a door. 2. The speech language training machine according to claim 1, wherein said speech recognition means performs continuous word speech recognition by utilizing syntactic information determined in advance in the content of the lesson.