JPH07134598A - System for outputting basic frequency of voice - Google Patents

Abstract

PURPOSE:To provide a system capable of adjusting an output state of a fundamental frequency pattern according to the change in the size of an output surface and the number of data and visually providing a clear result. CONSTITUTION:A voice waveform analysis part 12 extracts a basic frequency and an amplitude value at every prescribed vocal sound line section from a voice waveform, and stores it in an analysis data hold part 16 temporarily. A syllable time length normalization part 13 normalizes a time length of other syllable based on an analysis time length of one syllable, and comparison observes plural basic frequency patterns on the same time base. Although the normalized time length is led to a display position operation part 14 for deciding a display position on an output device 2, the size information of the output surface is considered at the time of the decision, and further, only the time length suitable to an output condition of a lower limit value, etc., of a display interval is extracted, and the display data are formed. Thus, the redundant data are eliminated, and the clear output result is obtained.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech prosody analysis technique in the field of speech generation and recognition, and particularly to the fundamental frequency of syllables having different utterance time lengths (hereinafter, time lengths) even for the same character string. The present invention relates to a fundamental frequency output system for speech, which allows clear and comparative observation of patterns on the same time axis.

[0002]

2. Description of the Related Art Accent of voice, intonation,
Stress, rhythm, and the like are elements other than phonological characteristics, and are expressed by prosodic features such as the fundamental frequency and amplitude value of speech, and syllable connection time. These are important not only for transmission of linguistic information such as meaning content but also for transmission of non-verbal information such as emotions.

Of these, the accent is used for distinguishing a plurality of words (syllables) having the same monophonic structure in terms of meaning or part of speech, but which prosodic feature is mainly used depends on the language. For example, a Japanese accent is called a pitch accent, and is represented exclusively by a temporal change pattern of a fundamental frequency (hereinafter, fundamental frequency pattern). Further, intonation indicates a syntactic structure and a distinction between a plain text and an interrogative sentence, but is expressed by a fundamental frequency pattern in many languages including Japanese.

As described above, since the fundamental frequency pattern of a voice is an important prosodic feature that is a basis for analyzing various information, in the field of voice generation and recognition, the fundamental frequency of an input voice waveform is used. It is required to extract patterns quickly and accurately. Therefore, it is generally practiced to obtain a basic frequency pattern of a voice for each predetermined analysis time length using an information processing device and output the basic frequency to a printer or a display device at predetermined intervals to obtain a basic frequency pattern.

By the way, it is well known that the same character string has different time lengths depending on the speaker or the utterance state at that time. Especially, it is remarkable in terms of syllables. When comparing and observing the fundamental frequency patterns of syllables having different time lengths on the graph surface of the time axis versus the frequency axis, for example, conventionally, the fundamental frequency for each syllable section is temporarily held and a plurality of The time axis of the pattern was modified to be the same so that only the differences in prosodic features appeared.

[0006]

However, in the conventional method, when a plurality of fundamental frequency patterns are displayed on the same graph surface, the intervals of printing of each pattern (including the case of display only, the same applies hereinafter) and Since the number of prints (the number of data) was fixed, when the size of the output surface on which the graph surface is printed, for example, the size of the printing paper surface or the number of data to be printed changes, the printed parts overlap and the difference in clear prosodic features becomes apparent. Was not obtained in some cases. Also, noise components may be mixed depending on the speaker or the utterance state, but in the past, since the fundamental frequency was detected for all input speech, this was reflected on the graph surface and the fundamental frequency pattern was clear. There was a case where it deteriorated.

The present invention was created in view of such a background, and an object thereof is to quickly obtain a visually clear fundamental frequency pattern regardless of the change in the size of the output surface or the utterance state. To provide a fundamental frequency output system.

[0008]

According to the present invention, the display position of each fundamental frequency is restricted by the size of the output surface, and for noise components other than voiced sound, the amplitude value is voiced. The object is achieved by utilizing the property of being lower and having no periodicity correlation.

That is, the first invention is a system for displaying a fundamental frequency pattern of a voice waveform of a syllable or a phonological sequence on a two-dimensional output surface including a time axis, and for each predetermined phonological sequence section from the voice waveform. At least one of a voice waveform analysis unit that extracts a fundamental frequency and an amplitude value, an analysis data holding unit that holds the extracted fundamental frequency and an amplitude value corresponding to the fundamental frequency as analysis data of the phoneme section, and at least the output surface. Output surface information holding means for holding dimension information in the time axis direction, display position determining means for determining the display position of each fundamental frequency on the output surface based on the dimension information in the time axis direction, and the output surface The output condition of the analysis data is stored, and the fundamental frequency and its display position that meet this output condition are extracted from the analysis data holding means and the display position determining means, and extracted. Providing the fundamental frequency output system of a sound and a display data generating means for generating display data to the output surface in accordance with the present frequency and the display position.

The second invention is a system for simultaneously outputting a fundamental frequency pattern of a speech waveform of a plurality of syllables or phoneme strings having different utterance time lengths to a two-dimensional output surface including a time axis. A voice waveform analysis means for extracting a fundamental frequency and an amplitude value for each predetermined phoneme sequence section from the waveform, and holding each extracted fundamental frequency and an amplitude value corresponding to the fundamental frequency as analysis data of the phoneme section. Analysis data holding means, utterance time length normalizing means for normalizing utterance time lengths of other syllables or phoneme strings based on the utterance time length of a particular one of the plurality of syllables or phoneme strings, and the output surface Output surface information holding means for holding at least dimension information in the time axis direction, and display position determination for determining the display position of each fundamental frequency on the output surface based on the dimension information in the time axis direction. And the output condition of the analysis data to the output surface is held, and the basic frequency and its display position that meet this output condition are extracted from the analysis data holding means and the display position determining means, and the extracted basic frequency and There is provided a fundamental frequency output system for voice, comprising: display data creating means for creating display data on the output surface based on a display position.

As the output conditions of the display data creating means, various ones can be set according to the application, but in order to more effectively solve the conventional problems, in the present invention, the individual output surfaces are individually selected. At least one of the lower limit of the data display interval, the lower limit of the amplitude value corresponding to the fundamental frequency, and the threshold value of the autocorrelation coefficient value of the fundamental frequency. I decided to judge.

The component having no periodicity on the time axis (noise component) has an autocorrelation coefficient value smaller than that of the effective voice component. Therefore, a threshold value is set in advance, the presence or absence of periodicity is detected by comparing the autocorrelation coefficient value with this threshold value, and when there is no periodicity, this can be determined as a noise component.

[0013]

In the first aspect of the invention, the fundamental frequency and the amplitude value for each phoneme sequence section extracted from the voice waveform by the voice waveform analyzing means are temporarily stored in the analysis data holding means. The output of the voice waveform analyzing means is also guided to the display position determining means for determining the display position on the output surface, and the dimension information of the output surface is taken into consideration in the determination. For example, as the size in the time axis direction increases, the individual display intervals also increase. The display data creating means holds an output condition including at least one of a lower limit value of an individual data display interval on the output surface, a lower limit value of an amplitude value corresponding to a fundamental frequency, and a threshold value of an autocorrelation coefficient value of the fundamental frequency. If the display interval determined above does not reach the lower limit value, or if the amplitude value of the fundamental frequency held in the analysis data holding means does not reach the lower limit value, the fundamental frequency or display position is determined to be an incompatible element. And ignore these elements. Further, when the value of the autocorrelation coefficient of the fundamental frequency is less than the threshold value, it can be seen that it is a noise component because the periodicity correlation cannot be obtained between adjacent frames. The fundamental frequency in this case is also determined to be an incompatible element and is ignored.

On the other hand, when the amplitude value of the fundamental frequency and the display interval exceed the respective lower limit values, it is determined as a compatible element and display data is created. This surely avoids multiple printing and also prevents the output of noise components.

A second invention is a configuration for simultaneously outputting a fundamental frequency pattern of a speech waveform of a plurality of syllables or phoneme strings having different utterance time lengths to a two-dimensional output surface including a time axis. In the invention of (1), a vocalization time length normalizing means is added. In this utterance time length normalizing means, the utterance time lengths of other syllables or phoneme strings are normalized with respect to the utterance time lengths of others. As a result, the analysis unit time length of each voice waveform becomes the same, and a plurality of prosodic features can be expressed on the same time axis. At that time, the output data and the output conditions are taken into consideration, and the display data of the individual fundamental frequencies corresponding to the respective voice waveforms are created, as in the case of the first invention.

[0016]

Embodiments of the present invention will now be described with reference to the drawings.

FIG. 1 is a block diagram of a voice fundamental frequency output system according to an embodiment of the present invention, in which the fundamental frequencies of a plurality of syllables or phoneme strings are printed on the same output surface. Is shown. In FIG. 1, reference numeral 1 is a fundamental frequency analyzer, and 2 is an output device such as a printer or a display.

The fundamental frequency analyzer 1 inputs a speech waveform and outputs a fundamental frequency pattern to the output device 2, and the speech waveform holding unit 1 for storing the speech waveform.
1, a voice waveform analysis unit 12 that calls up and analyzes a stored voice waveform at any time to obtain a fundamental frequency and its amplitude value, a syllable time length normalization unit 13 that normalizes the time length of a plurality of syllables, and an output device 2. A display position calculation unit 14 that calculates position information on the output surface, a display data creation unit 15 that creates coordinate data to be actually displayed on the output surface, and the voice waveform analysis unit 1 described above.
2. An analysis data holding unit 16 which stores the fundamental frequency and the amplitude value obtained in step 2 in association with each other, and a syllable section information holding section which holds syllabic section information of the voice created based on phoneme labeling corresponding to the voice waveform. The unit 17 includes an output surface information storage unit 18 that stores information about the dimensions of the output surface, and a print condition storage unit 19 that sets and stores conditions for actual printing.

Next, the processing operation of the fundamental frequency analyzer 1 having the above configuration will be specifically described.

The input voice waveform is stored in the voice waveform holding unit 1.
1, and is sent to the voice waveform analysis unit 12 at any time in order to extract the fundamental frequency and the amplitude value.
In the extraction of the fundamental frequency, a very short time is defined as an analysis unit time length (hereinafter, frame), and one fundamental frequency is determined for one frame by analyzing a waveform cut out by this frame. Here, for example, even if the same syllable “fune” has different time lengths, the number of frames and the number of fundamental frequencies also differ.

The voice waveform from the voice waveform holding unit 11 is analyzed by the voice waveform analysis unit 12 as described above, and at this time, based on the syllable section information of the voice waveform sent from the syllable section information holding unit 17. The target syllable. The fundamental frequency of each frame of the extracted syllable and the amplitude value at this time are sent to the analysis data holding unit 16 as analysis data and stored therein. Although not shown, a known coefficient value calculating means for calculating an autocorrelation coefficient value for each fundamental frequency is provided, and the calculation result is also stored in the analysis data holding unit 16 as analysis data. good.

In order to compare and observe syllables having various time lengths on the same time axis, the present embodiment uses the time length of one syllable as a reference and calculates and uses the ratio to the case of other syllables. This ratio calculation is performed by the syllable time length normalization unit 13.

The display position calculation unit 14 calculates the interval on the time axis to be actually printed using the ratio obtained by the syllable time length normalization unit 13. At that time, it is stored in the output surface information holding unit 18. Refer to the output surface information. This output surface information is
Specifically, it is setting information indicating the dimensions of the output surface of the output device 2, for example, the vertical and horizontal lengths of the two-dimensional printing paper surface. This set value can be arbitrarily changed according to the application. Now, assuming that the horizontal length (length in the time axis direction) is L, the fundamental frequencies of a plurality of syllables are set to the same value within the length L by the maximum value of the time axis by the display position calculation unit 4. Aligned. Further, the print interval is automatically changed according to the setting information of the output surface.

The display data creation unit 15 creates display data for each frame of the above-mentioned analysis data based on the output of the display position calculation unit 14 and the printing conditions stored in the printing condition holding unit 19.

The printing conditions are, for example, the lower limit value of the printing interval corresponding to the size of the output surface and the lower limit value of the amplitude value for each fundamental frequency, and the display data is created only for the fundamental frequency exceeding these lower limit values.

The dimensions of the output surface can be arbitrarily set as described above, and the printing interval of the fundamental frequency is automatically changed according to the set value. Usually, when the fundamental frequencies of a plurality of syllables are printed, they overlap each other in the time axis direction. Therefore, the lower limit value of the printing interval is determined by determining the maximum value of the number of printed sheets in consideration of the lateral length L. For example, if the maximum number of prints is 20 per 200 [ms] on the time axis, the lower limit of the print interval is uniquely determined. If there are frames that do not meet this lower limit, the individual prints will overlap, so the fundamental frequency is ignored as a nonconforming element, and printing is not performed.

Further, for example, the amplitude value of the noise component in the case of recorded voice or when uttered using a microphone or the like is smaller than the amplitude value of voiced sound. In particular, the noise component generated at the time of utterance has an extremely small autocorrelation coefficient value, and since the periodic correlation between adjacent frames cannot be obtained, when the above-mentioned autocorrelation count value is less than a predetermined threshold value, It is also possible to determine that it is a noise component. Therefore, in the present embodiment, the lower limit value of the amplitude value and the threshold value of the autocorrelation coefficient value for each fundamental frequency are set, and the fundamental frequency that is less than the lower limit value or the threshold value is determined to be a noise component and printing is not performed. did.

The display data created by the display data creating section 15 is output to the output device 2 and printed on the output surface.

FIG. 2 is a graph in which the analysis data of the fundamental frequencies of a plurality of phoneme strings are plotted on the printing paper surface of the time axis versus the frequency axis by a conventional method that does not consider the output surface information and the printing conditions. The sentence "fune made mu
"kaeni ikimasuyo" is decomposed into ten phoneme strings "fu" to "yo", and the same person is uttered in different ways (state a to state d) on the same time axis. The reference time length for normalization is based on the time length of the character string of condition a.

In this figure, the state a is a normal state, the state b is a laughing state, the state c is a crying state, and the state d is an angry state. As shown in FIG. 2, it is possible to compare fundamental frequency patterns of syllables (phoneme sequences) having different time lengths by normalizing the time lengths. However, the printed portions are superposed and the noise components mixed in each of the states a to d are plotted through the same processing, so that they are visually unclear.

On the other hand, the graph surface obtained in this embodiment is shown in FIG.

As is apparent from FIG. 3, the fundamental frequency and the amplitude value held in the analysis data holding unit 16, the set value of the size of the output surface stored in the output surface information holding unit 18, and the printing condition holding unit 19 According to the fundamental frequency output system of the present embodiment, which controls printing of the fundamental frequency corresponding to each syllable based on each output condition stored in, the redundant or unnecessary analysis data is omitted, and a visually clear output is obtained. It turns out that the result is obtained.

Although FIG. 2 and FIG. 3 are examples of voice waveforms uttered by the same person in various states, the same applies to the case of only one voice waveform or voice waveforms uttered by different speakers. Needless to say, the processing of is possible.

As described above, according to this embodiment, it is possible to analyze linguistic information or non-linguistic information more easily and accurately. Moreover, whether or not the lower limit value of the display interval has been reached is determined based on the calculated value obtained from the display position calculation unit 14, so that even if the size of the output surface changes, the optimum printing is performed by following this. The spacing is obtained.

Although the present embodiment is as described above, the present invention is not limited to the above embodiment, and other embodiments are possible without departing from the scope of the invention.

[0036]

As described above, according to the fundamental frequency output system for voice of the present invention, the interval of the display data is re-determined according to the size information of the output surface, and the interval is a predetermined output condition. If the lower limit of the interval is not reached, display data will not be created until the lower limit is exceeded, so even if the number of fundamental frequencies to be output changes or the dimensions of the output surface change. It has the effect of being displayed on the output surface at intervals exceeding the above lower limit.

Further, by including the lower limit value of the amplitude value of the fundamental frequency or the threshold value of the autocorrelation coefficient value as the output condition, when the amplitude value does not reach the lower limit value or the autocorrelation coefficient value of the fundamental frequency is the threshold value. Since the display data is not created even when the value is less than, there is an effect that the noise component is not displayed on the output surface.

As described above, according to the present invention, the manner of display can be easily adjusted according to the output conditions provided in the display data creating means, and a visually clear fundamental frequency pattern can be quickly obtained.

[Brief description of drawings]

FIG. 1 is a block diagram of an audio fundamental frequency output system according to an embodiment of the present invention.

FIG. 2 is an explanatory diagram showing an example of an output result of a fundamental frequency when output surface information and output conditions are not considered.

FIG. 3 is an explanatory diagram showing an example of an output result of a fundamental frequency output by the configuration of the present embodiment in consideration of output surface information, a lower limit value of a print interval, and a lower limit value of a print interval.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 fundamental frequency analyzer 11 voice waveform holding unit 12 voice waveform analyzing unit 13 syllable time length normalizing unit 14 display position calculation unit 15 display data creating unit 17 syllable section information holding unit 18 output surface information holding unit 19 printing condition holding unit 2 Output device having two-dimensional output surface

Claims (3)

[Claims] 1. A system for displaying a fundamental frequency pattern of a speech waveform of a syllable or a phonological sequence on a two-dimensional output surface including a time axis, the fundamental frequency and amplitude value of each prescribed phonological sequence section from the speech waveform. A voice waveform analyzing means for extracting, an analysis data holding means for holding the extracted fundamental frequency and an amplitude value corresponding to the fundamental frequency as analysis data of the phoneme section, and a dimension of at least the time axis direction of the output surface. Output surface information holding means for holding information, display position determining means for determining the display position on the output surface of each fundamental frequency based on the dimension information in the time axis direction, of the analysis data to the output surface The output frequency is held, and the fundamental frequency and its display position that meet this output condition are extracted from the analysis data holding means and the display position determining means, and the extracted fundamental frequency is extracted. And fundamental frequency output system sound and having a display data generating means for generating display data to the output surface based on the display position. 2. A system for simultaneously outputting a fundamental frequency pattern of a speech waveform of a plurality of syllables or phoneme strings having different utterance time lengths to a two-dimensional output surface including a time axis, wherein each of the speech waveforms has a predetermined frequency. A speech waveform analysis unit that extracts a fundamental frequency and an amplitude value for each phoneme sequence section, and an analysis data holding unit that holds each extracted fundamental frequency and an amplitude value corresponding to the fundamental frequency as analysis data of the phoneme section. A voicing time length normalizing means for normalizing voicing time lengths of other syllables or phonological strings based on the utterance time length of a particular one of the plurality of syllables or phonological strings, and at least the time axis direction of the output surface Output surface information holding means for holding the dimensional information, display position determining means for deciding the display position on the output surface of each fundamental frequency based on the dimensional information in the time axis direction, While holding the output condition of the analysis data to the output surface, the fundamental frequency and its display position that meet this output condition are extracted from the analysis data holding means and the display position determining means, and the extracted fundamental frequency and display position And a display data creating unit for creating display data on the output surface based on the above. 3. The audio fundamental frequency output system according to claim 1, wherein the display data creating unit corresponds to a lower limit value of individual data display intervals on an output surface and a fundamental frequency as the output condition. The fundamental frequency of the voice, which includes at least one of the lower limit value of the amplitude value and the threshold value of the autocorrelation coefficient value of the fundamental frequency, and which determines the fundamental frequency or the display position less than the lower limit value or the threshold value as a nonconforming element. Output system.