JP2006084630A - Infant's voice analysis system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To understand the reason why an infant is crying from the crying voice of the infant. <P>SOLUTION: An inputted speech signal is subjected to frequency analysis to find the frequency spectrum of the speech signal and it is estimated why the infant is crying is "sleepiness" when degrees q1, q2, q3, and q4 of variations between adjacent frequency bands among frequency bands obtained by dividing the found frequency spectrum are larger than a predetermined threshold or "hunger" when the degrees of variations between adjacent frequency bands are smaller than the predetermined threshold. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an infant voice analysis system that analyzes and displays an infant's crying voice and displays the psychological state of the infant, and more particularly to an infant voice analysis system that improves the estimation accuracy of "sleepy" and "hungry". About.

Conventionally, what is disclosed in Patent Document 1 is known as a voice analysis system for grasping the cause of crying of an infant from the cry of the infant. This system performs frequency analysis of an audio signal for one breath of an infant, calculates a feature value based on the frequency spectrum of the audio signal, and estimates the cause of the infant's crying based on the calculated feature value. The cause of the crying is estimated based on the presence / absence of periodicity in each band of the frequency spectrum of the audio signal and the frequency band with periodicity. More specifically, for example, when the frequency spectrum of the audio signal has a periodicity from a low frequency band to a high frequency band, the crying cause estimation unit estimates that the cause of crying is “hungry” If the frequency spectrum has periodicity in a low frequency band, it is estimated that the cause of crying is “sleepy”, and the frequency spectrum of the audio signal does not have periodicity, or the period changes over time In this case, the cause of crying is estimated to be “painful”.
Japanese Patent Laid-Open No. 2002-2785582, paragraphs 0013 to 0018, FIGS.

  In the above-described conventional method for estimating the cause of crying, how high the frequency band of the horizontal stripe pattern that appears when the frequency spectrum of the crying voice for one breath of an infant is displayed in a two-dimensional space can be detected. Is used to discriminate between “hungry” and “sleepy”.

  However, depending on the cry data, the above-described fringe component is often unclear, and it may be difficult to detect the fringe component in a high frequency band even when hungry.

  This invention is made in view of such a point, and it aims at providing the audio | voice analysis system of the infant which makes it possible to grasp | ascertain the cause of an infant's cry more accurately from the infant's cry.

  An infant voice analysis system according to the present invention includes an input means for inputting a voice signal of an infant, a voice analysis means for obtaining a frequency spectrum of the voice signal by frequency analysis of the input voice signal, and the frequency spectrum. When the degree of change between adjacent frequency bands when dividing into a plurality of frequency bands is greater than a predetermined threshold, it is estimated that the cause of crying is "sleepy", and the adjacent frequency unit When the degree of change in the value is smaller than a predetermined threshold value, a cause of crying estimation that estimates that the cause of crying is “hungry” and a cause of crying estimated by the cause of crying estimation are displayed. And a display means.

  The crying cause estimation unit preferably includes a plurality of different estimation rules for each age of the infant to be analyzed, and a frequency band and a threshold value used for estimation are specified for each estimation rule.

  For example, when the age of the infant to be analyzed is 3 to 4 months, the cause of crying cause estimation means pays attention to the degree of change between the frequency units on both sides of the frequency spectrum around 9 kHz. When the age of the infant to be analyzed is 5 to 12 months, attention is paid to the degree of change between the frequency units on both sides of the frequency spectrum centering around 3 kHz. Then, it is estimated whether the cause of crying is “sleepy” or “hungry”, based on whether or not the degree of change of interest exceeds or exceeds the predetermined threshold value.

  Further, the crying cause estimation means preferably detects the presence or absence of periodicity in each band of the frequency spectrum of the audio signal prior to the estimation process of the cause of crying according to the degree of change in the frequency band, and the frequency of the audio signal When the spectrum has periodicity continuously from the low frequency band to the high frequency band, it is estimated that the cause of crying is “hunger”.

  According to the present invention, if the degree of change between adjacent frequency bands when the frequency spectrum of the infant's audio signal is divided into a plurality of frequency bands is greater than a predetermined threshold, the cause of crying Is assumed to be “sleepy”, and if the degree of change in adjacent frequency units is smaller than a predetermined threshold, the cause of crying is estimated to be “hungry”, so the frequency spectrum Even when the fringe component based on the periodicity cannot be detected, “hungry” and “sleepy” can be distinguished, and the cause of crying of the infant can be estimated more accurately.

  Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a functional block diagram showing the configuration of an infant voice analysis system according to an embodiment of the present invention.

  This system includes a microphone 1 that captures an infant's cry as an audio signal, an A / D converter 2 that samples the audio signal captured by the microphone 1 at a predetermined sampling frequency, and performs analog / digital conversion. A speech analysis unit 3 that analyzes the speech signal sampled by the converter 2 and calculates a feature amount based on the frequency spectrum, and estimates the cause of crying based on the feature amount of the speech signal obtained by the speech analysis unit 3 The crying cause estimation unit 4 and an estimation result display unit 5 that displays an estimation result in the crying cause estimation unit 4 are configured.

  This system is realized by software, hardware, or both, and can take various forms depending on the installation location of the system. For example, (1) the microphone 1 is installed in the vicinity of an infant, the voice is collected there, and the voice analysis unit 3, the crying cause estimation unit 4 and the estimation result display unit 5 placed in another place are wired or wirelessly A form in which an audio signal is sent and analyzed, estimated and displayed in another place, (2) a form in which the whole is installed in the vicinity of the infant, (3) a voice signal is collected, analyzed and estimated in the vicinity of the infant, Although the form etc. which display an estimation result on the estimation result display part 5 installed in another place for an estimation result can be considered, it is not restricted to these forms in particular.

  Next, as a specific analysis / estimation method, an example of a method for classifying three types of states of hunger, sleepiness, and pain by frequency analysis will be described.

  First, an infant's cry is taken from the microphone 1 and digitized by the A / D converter 2. At this time, the sampling frequency is set to a high value of 30 kHz or more, preferably 40 kHz or more (for example, 44.1 kHz, etc.) so that aliasing noise does not enter in order to see frequency components of 15 kHz or more.

  The obtained digital data is supplied to the voice analysis unit 3. The voice analysis unit 3 can be realized by a signal processing device such as a personal computer, a microprocessor, a DSP and the like together with the crying cause estimation unit 4. As a function, the speech analysis unit 3 includes a breath extraction unit 31 and a frequency analysis / feature amount calculation unit 32. Including. First, an audio signal for one breath is cut out. That is, as shown in FIG. 2, since the infant's cry is intermittently generated in conjunction with the infant's breathing and becomes a signal in which the sound part and silence part for one breath are repeated, 31 cuts out an audio signal for each section, using a section in which a certain level of sound pressure level continues as a signal for one breath.

  Next, as shown in FIG. 3, the frequency analysis / feature amount calculation unit 32 extracts N small sections from the audio signal of the extracted section at a predetermined interval, and performs Fourier transform on each of the small sections. To obtain a frequency spectrum (power spectrum) for each small section and to calculate the feature amount. Note that since FFT (Fast Fourier Transform) is generally used as the Fourier transform method, it will be described below using other methods, but it goes without saying that other methods may be used.

  FIG. 4 is a diagram showing a frequency spectrum (power spectrum) at each time (N locations) and a sound spectrogram obtained by continuously obtaining the frequency spectrum and displaying the time on the horizontal axis and the frequency on the vertical axis. Causes of crying for infants include hunger, sleepiness, pain, loneliness, fear, discomfort, etc. Here, the sound spectrogram of crying regarding hunger and sleepiness is as follows.

  (1) On an empty stomach: When crying for one breath is cut out and the frequency spectrum is obtained for each of the N subsections in the cutout section, the obtained N frequency spectra (power spectra) are shown in FIG. Like (a), it becomes the substantially same periodic waveform from which a peak appears periodically from a low frequency to a high frequency. Therefore, when a sound spectrogram is obtained for a cry for one breath, horizontal stripes appear continuously from a low frequency to a high frequency. According to the experiments by the present inventors, in the sound spectrum of the hungry cry, (a) when horizontal stripes are detected at 15 kHz or higher, (b) when horizontal stripes are clearly detected at 11 kHz or higher, (c ) Horizontal stripes are slightly detected at 11 kHz or more, and there may be continuous horizontal stripes from 6 to 10 kHz.

  (2) When sleepy: When the cry for one breath is cut out and the frequency spectrum is obtained for each of the N subsections in the cutout section, the obtained N frequency spectra (power spectra) are shown in FIG. As shown in (b), the waveforms are substantially the same in which peaks periodically appear only in a low frequency band (about 0 to 6 kHz). Therefore, in the sound spectrogram of the cry for one breath, a horizontal stripe appears, but this appears only up to a low frequency band (about 0 to 6 kHz).

  In the above, “hungry” and “when sleepy” are identified from the horizontal stripes appearing in the sound spectrogram of the cry, but there are many pieces of data in which the stripe component is unclear in the cry data. Therefore, in this system, even when the fringe component cannot be detected, the accuracy of discrimination between “hungry” and “when sleepy” is improved by using another estimation method in combination. Specifically, attention is focused on how the magnitude of the frequency spectrum changes (whether there is a sudden decrease).

  That is, as shown in FIG. 5, the frequency spectrum is divided into a plurality of frequency bands in the frequency axis direction. Here, it is divided every 3 kHz, and 0 to 3 kHz, 3 to 6 kHz, 6 to 9 kHz, 9 to 12 kHz, and 12 kHz are set as frequency bands F1, F2, F3, F4, and F5, respectively. Now, assuming that the integral value of the spectrum | X (f) | of the frequency band Fi is Si, qi is defined as follows.

[Equation 1]
qi = Si / Si + 1
This qi is a feature amount indicating the degree of change between adjacent frequency bands Fi and Fi + 1.

  FIG. 6 shows the degree of change q1, q2, q3, q4 of the adjacent frequency bands when the infant is 3-4 months old when hungry and sleepy, and FIG. 7 shows the hungry state of the infant 5-5 months old. The degree q1, q2, q3, q4 of the change of the adjacent frequency band at the time of sleep and sleep is shown. As is clear from these figures, for infants 3 to 4 months old, q3 shows a high value of 5.9 when hungry and 8.4 when sleepy, and q1, q2, and q4 show 2.1 to The value was as low as 2.9. This indicates that the value of the frequency spectrum is drastically decreasing centering on 9 kHz, and indicates that the degree of decrease is greater when sleepy than when hungry. This tendency varies depending on the age, and for infants 5 to 12 months, q1 is as high as 6.5 when hungry and 8.9 when sleepy, and 2.4 to 3 for q2, q3, and q4 .8, a low value. This indicates that the value of the frequency spectrum is drastically decreasing around 3 kHz, and that the degree of decrease is greater when sleepy than when fasting.

  From the above, for infants 3 to 4 months of age, for example, q3 and q4, and for infants 5 to 12 months of age, q1 and q2, the values for fasting and sleepy If the numerical values between them are set as threshold values T1 and T2, respectively, the “hungry” state and the “sleepy” state can be distinguished.

  Note that the numerical values in FIGS. 6 and 7 are only average values, and some variation was observed in the actual samples. In consideration of the combined use with the detection of the horizontal stripe pattern described above, it is sufficient to consider the degree of change in the adjacent frequency band for the sample that could not be detected by the horizontal stripe pattern in practice. For infants 5 to 12 months old, the detection accuracy was highest when T1 = 6 and T2 = 3.

  Another difference between the cry sound spectrum when “hungry” and the cry sound spectrum when “sleepy” is the amplitude in the low frequency region. That is, the volume on an empty stomach is often higher than the volume on a sleepy day. Therefore, this point is also a preliminary determination factor.

Based on the above points, the frequency analysis / feature amount calculation unit 32 calculates the following as the feature amount.
a) N power spectrum values obtained by FFT at N locations.
b) Dispersion value in each frequency band of N power spectra.
c) Cepstrum obtained for each frequency band for each power spectrum.
d) The location of each peak for the periodicity detected in the power spectrum.

  Next, the crying cause estimation unit 4 estimates the cause of crying of the infant from the feature amount calculated by the voice analysis unit 3. That is, for the two types of hungry and sleepy, a rule considering the above-mentioned characteristics is created, and the cause of crying is estimated based on the rule.

  FIG. 8 is a flowchart of an estimation process used for analyzing a crying voice of an infant of 5 to 12 months.

  First, the horizontal stripe pattern that appears in the sound spectrogram of the cry is detected. That is, when horizontal stripes are detected at 15 kHz or higher (S1), when horizontal stripes are clearly detected at 11 kHz or higher (S2), horizontal stripes are slightly detected at 11 kHz or higher, and horizontal stripes continuously from 6 to 10 kHz. If it exists (S3), the estimation result is “hungry”. Further, when all of steps S1 to S3 are NO, whether or not the degree of change q1 and q2 of the adjacent frequency band is smaller than the threshold values T1 and T2, respectively, and the level of the low frequency band F1 is the threshold. It is checked whether or not it is larger than the value T3 (S4), and when all of these are satisfied, the estimation result is set to “hungry”. If any one of the conditions in step S4 is not satisfied, the estimation result is “sleepy”.

  In addition, the above is the estimation rule of the cry of 5-12 months old infants, and the estimation rule with respect to infants with different ages needs to be prepared separately for each age.

  These estimation results are displayed in the estimation result display unit 5 as characters, images, light, voice, and the like. Thereby, since it is possible to notify both the fact of the crying and the cause thereof to a childcare person who monitors the display unit 5 at a position apart from the infant, the childcare support is extremely effective. be able to.

  In the above embodiment, “hungry” and “sleepy” have been described, but “pain” and “sweet” can also be identified by the same criteria.

1 is a block diagram of an infant voice analysis system according to an embodiment of the present invention. It is a wave form diagram which shows the audio | voice signal at the time of crying of the infant input into the system, and its cutting-out method. It is a figure for demonstrating the continuous FFT in the same system. It is a graph which shows the power spectrum and sound spectrogram according to the cause of crying observed with the system. It is a graph which shows the frequency spectrum of a cry voice signal. It is the table | surface which showed the degree of the change of the adjacent frequency band of the same frequency spectrum for every frequency and every cause of crying. It is the table | surface which showed the degree of the change of the adjacent frequency band of the same frequency spectrum for every frequency and every cause of crying. It is a flowchart which shows the process of the crying cause estimation part in the same system.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Microphone, 2 ... A / D converter, 3 ... Speech analysis part, 4 ... Crying cause estimation part, 5 ... Estimation result display part, 31 ... Extraction part for one breath, 32 ... Frequency analysis and feature-value calculation part

Claims (5)

An input means for inputting an infant's voice signal;
A voice analysis means for analyzing the frequency of the input voice signal to obtain a frequency spectrum of the voice signal;
When the degree of change between adjacent frequency bands when dividing the frequency spectrum into a plurality of frequency bands is greater than a predetermined threshold, it is estimated that the cause of crying is "sleepy", When the degree of change in adjacent frequency units is smaller than a predetermined threshold value, crying cause estimation means for estimating that the cause of crying is “hungry”;
An infant voice analysis system comprising: display means for displaying a cause of crying estimated by the cause of crying estimation. The crying cause estimation means includes a plurality of different estimation rules for each age of infant to be analyzed, and a frequency band and a threshold value used for estimation are specified for each estimation rule. The speech analysis system according to claim 1.   When the age of the infant to be analyzed is 3 to 4 months, the crying cause estimation means determines the degree of change between the frequency units on both sides of the frequency spectrum centered around 9 kHz. The infant voice analysis system according to claim 1, wherein whether the cause of crying is "sleepy" or "hungry" is estimated based on whether the threshold value is exceeded or not exceeded.   When the age of the infant to be analyzed is 5 to 12 months, the crying cause estimation means determines the degree of change between the frequency units on both sides of the frequency spectrum centering around 3 kHz. The infant voice analysis system according to claim 1, wherein whether the cause of crying is "sleepy" or "hungry" is estimated based on whether the threshold value is exceeded or not exceeded. The crying cause estimation means detects the presence or absence of periodicity in each band of the frequency spectrum of the voice signal prior to the estimation process of the cause of crying according to the degree of change in the frequency band, and the frequency spectrum of the voice signal is low frequency. The infant voice analysis system according to claim 1, wherein the cause of crying is estimated to be “hunger” when the band has periodicity continuously from the band to the high frequency band.

Images