JPS61121077A - Enunciation training apparatus - 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] [Field of Industrial Application] The present invention provides vocal training in which the distance between the student's vocal pattern and the teacher's vocal pattern is calculated, and the intonation of the vocalization is evaluated based on the distance. Regarding the method.

[Prior Art and Problems to be Solved by the Invention] Most conventional vocal training methods primarily use texts and audio tapes.

However, in this first method, the quality of the student's vocalization and the difference from the model teacher's vocal pattern are dependent on the auditory function of the student, who uses his/her ears to judge.

For this reason, students may not be able to make accurate judgments, resulting in a problem that the learning effect is not improved.

As a second method, by combining a personal computer and a voice input device, voice power, fundamental frequency patterns, etc. are converted into visual information as evaluation features of accent and intonation, and the result is shown to students.

However, this method has a problem in that it is difficult for students to objectively judge the quality of their own vocalizations using this visual information.

[Means for solving problems]

According to the present invention, in a vocal training method in which an input voice is converted into a fundamental frequency time series pattern by a voice analysis circuit, and then converted into visual information by a visual information output circuit and displayed on a display input section, the voice analysis circuit A logarithmic conversion circuit for converting the basic frequency time series pattern into a logarithm and an unvoiced sound part interpolation circuit for linearly interpolating the unvoiced part of the pattern after the logarithmic conversion are sequentially connected to the output side of the unit, and the output of the unvoiced sound part interpolation circuit is connected to the above-mentioned visual interface. An average value calculation circuit that is input to the information output circuit and is connected to the output side of the unvoiced sound part interpolation circuit to calculate the average value of the basic frequency time series pattern after the logarithmic transformation interpolation, and an average value calculation circuit that calculates the amount of displacement with respect to the average value and after the logarithmic transformation interpolation. A correction conversion circuit that converts into a fundamental frequency displacement amount time series pattern is connected in sequence, and a switching circuit that is switched by a signal from the display input section and the teacher's displacement amount time series pattern are stored on the output side of the correction conversion circuit. A dictionary memory for storing the student's displacement time-series pattern; a matching circuit for calculating the distance that is the difference between the teacher and student patterns; By providing an evaluation circuit that evaluates the audio and sends it to the display input section, the switching circuit is switched in response to an instruction from the display input section to store the teacher pattern and student pattern in both memories and perform the matching. There is provided a vocal training method characterized in that the distance is calculated by a circuit and the evaluation sent from the evaluation circuit is displayed on the display input section.

[For production]

The method of the present invention logarithmically transforms the fundamental frequency time series pattern, which is a characteristic parameter of input speech, in order to absorb fluctuations in the pitch of the voice based on the individuality of a speaker, and also interpolates unvoiced sound intervals. Conversion Converts the fundamental frequency time series pattern into a displacement pattern from the average value to absorb sound pitch fluctuations between multiple speakers, and stores the displacement pattern in separate memories by switching between the teacher pattern and the student pattern. Since the displacement pattern difference between the two can be compared as a distance, it becomes possible to objectively evaluate the student pattern with respect to the teacher pattern, and the learning motivation of the student who performs vocal training is improved.

〔Example〕

Hereinafter, the present invention will be explained by way of examples with reference to the accompanying drawings.

FIG. 1 is a diagram showing the configuration of an apparatus for implementing the vocal training method according to the present invention.

In the above device, the student's voice and the teacher's voice input from the microphone 1 are stored in the temporary memory 11 and the latter in the dictionary memory 14 by a switching circuit 1o, respectively.
Based on instructions from the display input section 7, a matching circuit 12 compares the two voices, and the result is evaluated by an evaluation circuit 15 and displayed on the display input section 7.

The microphone 1 is an input medium for inputting the student's or teacher's voice, the A/D converter 2 is a device for converting the input voice from analog to digital, and the voice analysis circuit 3 extracts a fundamental frequency pattern from the digital input voice. circuit,
The logarithmic conversion circuit 4 is a circuit that converts the fundamental frequency pattern into a logarithmic scale, the unvoiced sound part interpolation circuit 5 is a circuit that linearly interpolates the unvoiced sound section of the fundamental frequency gap, and the visual information output circuit 6 is a circuit that visually converts the sound after the interpolation. A circuit for converting into information, a display input unit 7 for displaying the audio and inputting control signals to the switching circuit 10 and memory instruction circuit 13, an average value calculation circuit 8 for calculating the average value of fundamental frequencies, A correction conversion circuit 9 is a circuit that converts the average frequency into a time-series pattern of displacement amount with respect to the average fundamental frequency, and a switching circuit 10 is a circuit that converts the average frequency into a time-series pattern of displacement amount with respect to the average fundamental frequency, and a switching circuit 10 converts the teacher's voice to the dictionary memory 14 side in the case of a student voice according to the command from the display input section 7. are further switched to the memory 11 side, the temporary memory 11 is a memory for temporarily storing the displacement basic frequency pattern of the student voice for matching with that of the teacher voice, and the matching circuit 12 is a memory that stores the student displacement of the temporary memory 11. Volume pattern and dictionary memory 1
The memory instruction circuit 13, which is a circuit that reads the teacher displacement amount pattern No. 4 and calculates the distance between both patterns after time normalization, stores the teacher pattern in the dictionary memory 14 via the switching circuit 10 according to instructions from the display input section 7, or A circuit for inputting the stored teacher pattern to the matching circuit 12, dictionary memory 1
4 is a memory for registering teacher patterns and using them as model patterns for student patterns; evaluation circuit 1;
Reference numeral 5 denotes a circuit that evaluates the distance between the teacher pattern and the student pattern calculated by the matching circuit 12 and sends the evaluation result to the display input section 7.

The operation of the apparatus shown in FIG. 1 having the above configuration will be explained below with reference to FIGS. 2 to 4.

It is assumed that the voice input from the microphone 1 is uttered according to a predetermined text, and the input voice is digitally converted by the ^/mouth converter 2 in the next stage.

Next, the audio analysis circuit 3 extracts a time-series pattern of fundamental frequencies from the digitally converted input audio.

FIG. 2 is a graph showing an example of a time-series pattern of fundamental frequencies extracted from the voice uttered by an adult male speaker as renjoy J.

That is, as shown in Figure 2 (B), if we draw a graph with time t on the horizontal axis and fundamental frequency f on the vertical axis, we get "e
", and two peaks can be seen at rnJ, and a time-series pattern is obtained in which there is a break between rnJ and "j", the maximum is at roJ, and the lowest is at "y." This figure 2 (B) is shown above [
The audio waveform corresponding to enjoyJ is shown in Figure 2 (A
)become that way. Next, the fundamental frequency time series pattern outputted from the speech analysis circuit 3 is converted into the logarithmic conversion circuit 4 of the next stage.
Convert to logarithmic scale.

Now, if the fundamental frequency pattern obtained in circuit 3 is F, then the frequency of each audio element is fl r2...fn
As F=f1. f2. =−”−fn −=−−−−−
-(1) Therefore, if equation (1) is transformed logarithmically, log F=log 11, log f2-1og
fn -42).

The reason for logarithmic transformation in this way is that when a fundamental frequency pattern is expressed on a logarithmic scale as shown in equation (2), the shape of the pattern remains constant regardless of the speaker's characteristic voice pitch. This is because the voice becomes calmer, thereby absorbing fluctuations in the pitch of the voice based on the personality of the speaker, and making the evaluation more objective.

This logarithmically transformed fundamental frequency time series pattern log F
The unvoiced sound section is linearly interpolated by the unvoiced sound section interpolation circuit 5.

That is, as can be seen from the graph of the logarithmically transformed fundamental frequency time series pattern shown in FIG. 3, the portion between rnJ and "j" is missing. This is because in the speech analysis circuit 3, the fundamental frequency is observed only in the dominant parts of voiced sounds accompanied by band vibration, and is observed in unvoiced parts and voiceless parts, and parts that are voiced sounds but whose characteristics are not clear, such as the rnJ mentioned above. This is because it is difficult to extract in cases such as between and "j".

Therefore, the circuit 5 performs linear interpolation on the silent section L (here, the section where the fundamental frequency is difficult to be observed is collectively referred to as the unvoiced section) as shown by the broken line.

This linearly interpolated logarithmically transformed fundamental frequency time series pattern is on the other hand inputted to the average value calculation circuit 8 to calculate the average fundamental frequency fM.

That is, using this fM, the correction conversion circuit 9 calculates the average value f
Find the displacement time series pattern for M.

Now, if the displacement pattern is ΔF, ΔF=1og fl-fM, log f2-fM.

...log f m −f M −・−−(
41 This equation (4) is illustrated using a graph in the fourth
In this figure, the points indicated by arrows represent each term on the right side of equation (4).

The reason for performing this conversion is that it is possible to absorb variations in the pitch of voices between speakers and to emphasize the shape of the logarithmically converted fundamental frequency time series pattern itself (FIG. 4).

On the other hand, the interpolated pattern output from the unvoiced part interpolation circuit 5 is input to the visual information output circuit 6 (FIG. 3), where it is converted into visual information and displayed on the display of the display input section 7. (OUT).

By the way, this display input section 7 also accepts switching control of the switching circuit 1o, and if the voice input to the microphone I and subsequently converted by the correction conversion circuit 9 is that of the teacher, the switch of the switching circuit 10 is activated. The dictionary memory 14 (11) is switched. At the same time, the displacement amount time series pattern that has passed through the correction conversion circuit 9 is stored in the dictionary memory 14 via the memory instruction circuit 13 in response to a signal from the input section 7.

In the case of a student's voice coming from the microphone 1, the switch of the switching circuit 1o is temporarily switched to the memory 1 side by the signal from the input section 7, and the student's displacement time series pattern (Fig. 4) is stored in the memory 1. Ru. and memory instruction circuit 1
3, the teacher pattern in the dictionary memory 4 and the student pattern in the temporary memory 1 are read into the machine notching circuit 12. Both read patterns are subjected to distance calculation after time normalization.

The distance in this case is calculated as follows.

Let each term on the right side of equation (4) be △flsΔ”2...Δf
If n is a teacher pattern, a subscript 1 is added to the right shoulder, and a student pattern is expressed by a subscript 2 to the right shoulder, as shown below.

When time normalization is performed on equations (5) and (6), ΔG=Δgt, 6g2...ΔgN −
Song - (5)'ΔG = Δgt, 6g2...Δg
N - (6)'.

(51', (In formula 61', N is the number of normalized divisions,
Δg is the average value of Δf in each section divided according to N.

The difference between the teacher pattern and the student pattern expressed by the equation (51', (6)' is defined as the distance between the two.

The distance d is input to the evaluation circuit 15 and used as an evaluation amount of vocal intonation, and the distance d is sent to the display input section 7 in accordance with a predetermined evaluation standard.

By displaying this standard on the display (OUT>), it is possible to inform the student of the quality of the vocalization from the device side.

〔Effect of the invention〕

According to the present invention, a fundamental frequency time series pattern, which is a characteristic parameter of input speech, is logarithmically converted to absorb fluctuations in voice pitch based on the individuality of a speaker, and unvoiced sound intervals are interpolated, and after the interpolation, Logarithmic conversion Converts the fundamental frequency time series pattern into a displacement pattern from the average value in order to absorb sound pitch fluctuations between multiple speakers, and switches the displacement pattern between the teacher pattern and the student pattern and stores them in separate memories. Store both of them,
Since the displacement pattern difference can be compared as a distance, it becomes possible to objectively evaluate the raw i-running pattern with respect to the teacher's pattern, thereby increasing the student's motivation to learn vocal training.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of an apparatus for carrying out the method of the present invention, and FIGS. 2(A) and 2(B) to FIG. 4 are explanatory views of the operation of the apparatus shown in FIG. 1. ■...Microphone, 2...A/D converter,
3... Voice analysis circuit, 4... Logarithmic conversion circuit, 5
...Unvoiced sound part interpolation circuit, 6...Visual information output circuit, 7...Display input section, 8...Average value calculation circuit, 9...Correction conversion circuit, 1o...Switching circuit, I
L...Temporary memory, 12-way matching circuit, 1
3...Memory instruction circuit, 14...Dictionary memory, 15...Evaluation circuit.

Claims (1)

[Claims] In a vocal training method in which an input voice is converted into a basic frequency time series pattern by a voice analysis circuit, and then converted into visual information by a visual information output circuit and displayed on a display input section, the basic frequency pattern is displayed on the output side of the voice analysis circuit. A logarithmic conversion circuit that converts the frequency time series pattern into a logarithm and an unvoiced sound part interpolation circuit that linearly interpolates the unvoiced part of the pattern after the logarithmic transformation are connected in sequence, and the output of the unvoiced sound part interpolation circuit is inputted to the visual information output circuit. Further, on the output side of the unvoiced sound part interpolation circuit, there is an average value calculation circuit that calculates the average value of the fundamental frequency time series pattern after logarithmic transformation interpolation, and an average value calculation circuit that calculates the average value of the fundamental frequency time series pattern after logarithmic transformation interpolation, and an average value calculation circuit that calculates the displacement amount with respect to the average value and generates the fundamental frequency displacement amount time series after logarithmic transformation interpolation. A correction conversion circuit for converting into a pattern is connected in sequence, and furthermore, on the output side of the correction conversion circuit, a switching circuit that is switched by a signal from the display input section, a dictionary memory for storing the teacher's displacement time series pattern, and a student's A temporary memory that stores the displacement time series pattern; a matching circuit that calculates a distance that is the difference between the teacher and student patterns; and a matching circuit that evaluates the student's input voice with respect to the teacher's input voice based on the distance. By providing an evaluation circuit that sends data to the display input section, the switching circuit is switched in response to an instruction from the display input section, the teacher pattern and student pattern are stored in both memories, and the distance is calculated by the matching circuit. A vocal training method characterized in that the evaluation sent from the evaluation circuit is displayed on the display input section.