JPS60201376A - Enunciation training machine - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION (a) Technical Field of the Invention The present invention is based on the voice power and pitch information of the utterances of 9 vocal trainees, and determines the differences from the sample utterances by the length, strength,
The present invention relates to a vocal training machine that performs vocal training by detecting and displaying pitch and the presence or absence of insertion of unnecessary vowels.

(b) Technical Background If you want to teach yourself how to pronounce a foreign language, etc., one method is to play back the voice of a standard person recorded on magnetic tape, etc., and train by listening to the pronunciation sounds and pronouncing them. is commonly practiced. but.

In this method, the trainee judges subjectively whether the pronunciation is correct or not, so even a wrong pronunciation may be judged as correct depending on the trainee's own judgment, and there is no true correct pronunciation. There was a problem that they were trained without being trained.

Therefore, there has been a demand for the development of a vocal training machine that has the function of objectively determining whether or not a vocalization is correct, and that enables efficient vocal training.

(C) Conventional technology and problems Next, a conventional voice training machine will be explained.

As mentioned in section (b), conventional vocal training through general self-study involves playing back the teacher's voice using a magnetic tape, etc., and practicing vocalization by listening to this, but it is difficult to objectively determine whether one utterance is correct or not. The effect was weak because no judgment was made. Therefore, Japanese Patent Application Laid-Open No. 57-577 proposed a method of comparing and evaluating the voice spectrum of a sample and the voice uttered by a trainee using voice analysis technology.
-44178 "Voice training device" and JP-A-58-17
An invention application was filed for the ``268Or pronunciation practice device''.

However, since these inventions evaluate the comparison between the utterances of a sample and the utterances of a vocal practitioner by calculating the degree of similarity, what are the specific points for a vocal practitioner? It was difficult to know exactly what to pay attention to and correct the pronunciation (therefore, the effect of practice was also weak).

(d) Purpose of the Invention The purpose of the present invention is to provide a new vocal training device that eliminates the above-mentioned drawbacks, and in particular, quantitatively analyzes the differences between the vocalizations of a sample and the vocalizations of a vocal practitioner. The purpose of this invention is to realize a voice training machine that enables more effective voice training by specifically and accurately presenting the differences between the two.

(e) Structure of the Invention The present invention provides a voice analysis means for extracting the power and pinch of the voice of a sample and the voice of a trainee, and the voice of the sample extracted by the voice analysis means. a feature analysis means for quantitatively analyzing the difference between the sample's vocalization and the trainee's vocalization; By providing a display means for displaying the sample's utterances and specifically and accurately presenting the differences between the utterances of the sample and the utterances of the trainee.

This can be achieved by using a voice training machine that enables more effective voice training.

(f) Examples of the invention The present invention will be explained below with reference to the drawings.

FIG. 1 shows an embodiment of the vocal training machine according to the present invention.

Fig. 2 is an example of the feature analysis unit according to the present invention, Fig. 3 is a diagram showing differences in accents, (A) is a diagram showing the pronunciation of a word in terms of power and pitch, and (B) is Figure 4 shows the quantitative analysis of the differences between the pronunciation of the sample and the pronunciation of the trainee.
(B) is a diagram showing differences in vowel lengths, Figure 5 is a diagram showing differences in vocalizations in conversation, (A) is a diagram showing power and pitch, and (B) is a diagram showing the differences in utterances by DP matching. (C) is a diagram showing the quantitative analysis, and Figure 6 shows the quantification parameter table.

In the figure, 1 is a microphone, 2 is an audio storage section, 3 is a playback section, 4 is an input switching section, and 5 is a speaker.

6 is an analog/digital conversion section (hereinafter referred to as ^/D conversion section), 7 is an analysis section, and 8 is a feature analysis section.

Reference numeral 9 denotes a feature storage section, IO a comparison evaluation section, 11 a judgment condition storage section, 12 a display section, 13 a vocalization time detection section, 14 a vocalization section detection section, and 15 a power detection section.

16 is a peak detection section, 17 is a pitch detection section, and 18 is a pitch amplitude detection section.

In addition, TM is the vocalization time, T1"Tn is the vocalization interval, t!1
~Rn is the power amount* Pl-Pn is the peak value, PTI
xPTn is the pitch, PTmax is the maximum pitch,
PTmin is the minimum pitch, and (1) is the sample's vocalization (
teacher's vocalization), +21°(2)' is the trainee's vocalization (
each student's vocalization).

In this embodiment, a microphone 1.
A voice storage unit 2 in which the sample's vocalizations are stored. A playback unit 3 that plays back the vocalizations of the sample in the voice storage unit 2. an input switching unit 4 that switches input between the vocalization from the microphone l and the reproduced sound from the reproduction unit 3; a speaker 5 that reproduces the vocalization specified by the input switching unit 4 as a sound; An A/D converter 6 that converts the voiced sound (analog signal) specified by the input switching unit 4 into a digital signal. Analysis unit 71 that analyzes and extracts the power and pinch of the vocalization converted into a digital signal
A feature analysis unit 8 that analyzes the power and Pisochi waveform analyzed and extracted by the analysis unit 7 to extract quantitative features. A feature storage unit 9 that stores quantitative features extracted by the feature analysis unit 8. A comparative evaluation section 10 that compares and judges the features stored in the feature storage section 9 and the features newly output from the feature analysis section 8 according to the conditions in the judgment condition storage section 11. Judgment condition storage unit 1 that stores certain conditions for comparison and judgment in the comparison evaluation unit IO
1. It consists of a display section 12 that displays the output of the comparison evaluation section 10.

The feature analysis section 8 shown in FIG. 2 also includes a vocalization time detection section 13. which detects the vocalization time TM from the power waveform. Vocalization section detection unit 1 similarly detects the vocalization section T1xTn from the power waveform.
4. Power waveform and section signal T of utterance section detection unit 14 output
A power detection unit 15.1 detects powers ff1E1 to En based on ff1E1 to Tn. Similarly, the power waveform and vocalization section detection unit 14
The peak value P1 to Pr+ is determined by the output interval signal TI=Tn.
The pitch PTI is determined by the pitch waveform of the beak detector 16 which detects
~Pitch detection section 17 for detecting PTn. It consists of a pitch amplitude detection section that detects pitch amplitude (PTmax-PTmin) from a pitch waveform.

For example, when studying English pronunciation by yourself, there are two words that show the difference in pronunciation between Japanese and English speakers:
As shown in Figure (A), there are differences in the vowel parts that are accented; or, as shown in Figure 4 (A), the insertion of unnecessary vowels by Japanese people pronouncing them in katakana;
There are differences in the length of the vowel part as shown in B). Furthermore, in conversation, as shown in FIG. It can be seen that there are large differences in the locations where voicing is strong (peak values P1 to Pn, power El to fin distribution for each voicing section T1 to Tn), changes in pitch PTI-PTn, etc.

Therefore, the power E1 to Bn due to the correct utterance of the sample and the temporal change in pitch PTI = PTn and the power [! It can be seen that comparing the temporal changes of pitches PTI to PTn with those of pitches PTI to PTn, analyzing the differences, and quantitatively pointing out and displaying them is very effective in increasing the effectiveness of language training.

Next, the operation of this embodiment shown in FIGS. 1 and 2 will be explained.

The utterances of the sample are recorded in the voice storage unit 2, which is a tape recorder or compact disk, etc., and the utterances of the trainee are input through the microphone 1, and by switching this with the input switching unit 4, the utterances of both parties can be changed. Voice sounds are input separately, and the input vocal sounds can be heard through the speaker 5. The input vocal sound is converted to A/D by the A/D converter 6.
After the conversion, the voice is analyzed by the second analysis unit 7, and the power [! 1
~Un and pitch PTI~PTn waveforms are extracted.

The waveforms of this power 1!1-Un and pitch PTI ``PTn'' are analyzed by the feature analysis section 8 to determine the utterance time TM and the peak value P1''P.
n, length of vocalization section T1=Tn, pitch width (PTmax
-PTa+in), etc. are calculated, and the criteria conditions for each word or conversation sentence that are stored in advance to determine whether or not the utterance is correct are extracted from the criteria storage unit 11 and compared and evaluated in the comparison evaluation unit 10. Then, the evaluation result is displayed on the display unit 12. Furthermore, the judgment condition storage unit 11 has, for example, Tl>T3>T2.
゜Pi>P3. PTmax-PT+++in >40%
, 0.1 <El<0.2.0°8 <82<0.9
Conditions such as Further, the parameters calculated and quantified by the feature analysis section 8 are displayed as parameters as shown in FIG.

FIG. 3(B) shows an example of an analysis result based on this example.

That is, the difference between the pronunciation of the word ro 'CLOCK J by the prototype and the pronunciation by the trainee is determined by the utterance interval T1.
T2, is divided into silent intervals TI2, and each interval T1. T
Peak value PI of power E1, 1!2 for every 2, P2 or power 1!1. f! 2 is calculated according to a predetermined calculation method. Also pitch PT1. Similarly for PT2, the pitch PTI, the peak value of PT2, and the pitch change width (
PTmax - PTmin), etc., and compare the differences as shown in FIG. 6 to perform respective evaluations.

On the other hand, the analysis of conversational utterances is performed quantitatively as shown in FIG. 5(C). However, the utterance section Tl of the conversation
Since the number of Tn is not clearly the same as in words, the utterances of the sample and the utterances of the trainee are compared by DP matching (in the field of operation research) on the speech spectrum as shown in Figure 5 (B). , a procedure for optimizing multi-step problems in which there are many conclusions at each stage of the process), and then quantitatively analyze the differences in the same way as in the case of words.

Figure 6 quantifies the analysis results shown in Figure 3 (B) (however,
The displayed values of peak values Pl to Pn and pitches PTI to PTn are relative values when the maximum value is 1, and the power E! 1-E
The displayed value of n is the total power E = 8171-1!2
+...+En=1, the power of each section [! 1~
It is expressed as a ratio of Bn).

The evaluation method in the comparative evaluation section 10 is performed as follows. However, those with a dash (') symbol are for use by trainees.

[l]: Evaluation of vocalization time TM ■Difference in total vocalization time... (T1+72+TI2)
-(Tl'+T2'+T12') ■Balance of voiced section... (Tl-Tl'), (T2-T2
') ... [2]: Power El~l! Peak value of n P1”Evaluation of Pn ■Position of maximum pinpoint value■Balance of peak values... (PI-PI'), (P
2-P2') ... [3]: Distribution of power E1=En■ Balance of power distribution... (EI-E1'), (
Bit! 2') ... [4]: Evaluation of pitch PTI ~ PTn ■ Maximum pitch position ■ Pitch balance... (PTI - PTI '
), (PT2-PT2')... ■Pitch change width...(PTmax -PTmin
) −(PTmax′−PTmin′).

(g) Effects of the Invention According to the present invention as described above, it is possible to quantify and specifically display the differences between the utterances of the sample and the utterances of the trainee, so that vocal training can be carried out efficiently. This has the effect of providing a voice training machine that is possible and has a great training effect.

[Brief explanation of drawings]

FIG. 1 shows an embodiment of the vocal training machine according to the present invention. FIG. 2 is an embodiment of the feature analysis unit according to the present invention, FIG. 3 is a diagram showing differences in accents, and FIG. 4 is a diagram showing differences between a sample's vocalizations and a trainee's vocalizations. FIG. 5 is a diagram showing differences in vocalizations in conversation, and FIG. 6 is a quantification parameter table. In the diagram, 1 is the microphone, 2 is the audio storage unit, 3
4 is a playback section, 4 is an input switching section, and 5 is a speaker. 6 is an A/D conversion unit, 7 is an analysis unit, and 8 is a feature analysis unit. 9 is a feature storage section, 10 is a comparison evaluation section, 11 is a judgment condition storage section, 12 is a display section, 13 is a vocalization time detection section, 14 is a vocalization section detection section, and 15 is a power detection section. 16 is a peak detection section, 17 is a pitch detection section, and 18 is a pitch amplitude detection section. Mine 3 songs (8) things srf1

Claims (1)

[Scope of Claims] +1) Voice analysis means for extracting the power and pitch of the voice of the sample and the voice of the trainee, and the voice of the sample extracted by the voice analysis means and the training. a feature analysis means for quantitatively analyzing the difference between the sample's vocalization and the trainee's vocalization, and a display that visually makes clear the difference between the sample's vocalization obtained by the voice analysis means and the trainee's vocalization. A vocal training machine characterized by being provided with a display means for displaying. (2) The feature analysis means performs quantitative analysis after making a correspondence between the utterances of the sample's utterances and the trainee's utterances. The vocal training machine described in paragraph 1. (3) The feature analysis means includes first means for detecting voiced sound sections from the power waveform of the voice, and second means for detecting the number and length of the voiced sound sections in the power waveform; and third means for calculating the peak value, power distribution, and pitch amplitude width of each voiced sound section detected by the first means. The vocal training machine described in Section 2.