KR102110285B1 - Hearing test method and apparatus based on speech dynamic range by frequency band - Google Patents

Abstract

A hearing test method based on calculating speech energy for each frequency according to an embodiment of the present invention includes calculating a speech dynamic range for each frequency of a test file, and a band audible following a band audible function for each frequency based on the calculated speech dynamic range Calculating a value, calculating a frequency importance for the same frequency from which the band audible value is calculated, and calculating a speech intelligibility index based on the band audible value and the frequency importance, in the test file The operation of calculating the speech dynamic range for each frequency is to remove silence in the data from the test file, calculate the average effective value level of the data from which the silence has been removed, and then normalize each recording file to a certain dB SPL. , Calculating signal envelopes in a plurality of critical bands, curveting the signal envelopes, songs It may include sub-sampling the linearized signal envelope and performing unit conversion, and calculating the dynamic range of speech by converting the cumulative signal envelope distribution level into a decibel signal envelope histogram.

Description

Hearing test method and device based on the calculation of speech energy by frequency {HEARING TEST METHOD AND APPARATUS BASED ON SPEECH DYNAMIC RANGE BY FREQUENCY BAND}

The present invention relates to a hearing test method and apparatus based on calculating speech energy for each frequency. More specifically, the method and apparatus for calculating speech energy information, which is information required for calculating the speech intelligibility index of the present invention, for a plurality of frequency bands, and calculating the speech intelligibility index by using the calculated speech energy for each frequency band It is about. Based on this, various embodiments of the present invention can more accurately measure the hearing ability of the hearing test subject.

The model of articulation theory has been proposed to predict the extent to which speech is accurately transmitted through various communication devices under electroacoustic conditions. In particular, the articulation theory model evaluated the quality of speech delivered using the frequency importance function of speech and the audible speech cues. In the auditory science, prediction of some components of the articulation theory model is speech. Was used to develop the model. The initial speech recognition prediction model was developed to quantify the relationship between audible speech signals and speech intelligibility, and was used as the name of the articulation index, but was later renamed as Speech Intelligibility Index (SII).

 The speech intelligibility index is mainly used to predict the prognosis of the hearing aid before and after wearing it through a change in speech perception. As a method for calculating the speech intelligibility index, the count-the-dot method proposed by Mueller & Killion is widely used. The Count ?? the Dot method can calculate a speech intelligibility index (SII) by displaying a gray square and a plurality of dots as shown in FIG. 1.

More specifically, the horizontal axis in FIG. 1 represents the frequency of the input sound, and the vertical axis represents the intensity (level) of the input sound. In addition, the colored area indicated in FIG. 1 usually represents speech (speech) energy distribution based on speech intensity, and the dots displayed inside speech energy indicate importance by frequency. In addition, a line crossing the colored area indicates the subject's hearing threshold.

As shown in FIG. 1, the subject's hearing threshold is 10 dB at 250 Hz, and the speech energy distribution of the corresponding frequency is 10-40 dB, so that the subject can hear all sounds. In addition, the subject has a sharply bad hearing threshold starting from 2K, and in 4K, one of the lowest points is heard and nothing else is heard.

Also, according to FIG. 1, there are 100 points in the speech energy region, and 3 points corresponding to 250 Hz are shown. This means that the total point means 100%, and the three points belonging to 250 Hz have a significance level of 3%. In the 2K frequency band, 11 dots are displayed and it means that there is 11% importance, so it can be seen that the 2KHz band is a more important frequency band than the 250Hz band. In addition, by checking the point that exists in the lower part of the subject's audible threshold line (the part having a value greater than the audible threshold dB), it is possible to check the ratio that the target can hear among the entire audible area. The total number of points is 76 Therefore, it can be seen that the subject can hear 76% of the entire audible area.

This existing method (Count-the Dot) has the advantage of being a simple and simple method, but it does not take into account the actual speech energy distribution (collectively, the dynamic range is specified as 30 dB) and the importance of frequency only. Accurate data calculation is difficult depending on the displayed point, and there is a problem that an error may occur.

On the other hand, prior art related to this is Patent Publication No. 10-2015-0129661 (hearing test and hearing evaluation device).

An embodiment of the present invention is designed to solve a problem that may occur when calculating a speech intelligibility index by the conventional count-the dot method, and the frequency importance according to the application of the band audible value for each frequency band and the properties of the test file The purpose is to calculate a more accurate speech intelligibility index by performing the application.

A hearing test method based on calculating speech energy for each frequency according to an embodiment of the present invention includes calculating a speech dynamic range for each frequency of a test file, and a band audible following a band audible function for each frequency based on the calculated speech dynamic range Calculating a value, calculating a frequency importance for the same frequency from which the band audible value is calculated, and calculating a speech intelligibility index based on the band audible value and the frequency importance, in the test file The operation of calculating the speech dynamic range for each frequency is to remove silence in the data from the test file, calculate the average effective value level of the data from which the silence has been removed, and then normalize each recording file to a certain dB SPL. , Calculating signal envelopes in a plurality of critical bands, curveting the signal envelopes, songs It may include sub-sampling the linearized signal envelope and performing unit conversion, and calculating the dynamic range of speech by converting the cumulative signal envelope distribution level into a decibel signal envelope histogram.

Since various embodiments of the present invention calculate test results based on speech energy for each frequency, there is an effect of more accurately measuring speech recognition ability of a hearing test subject.

1 is a view showing a conventional method for calculating the speech intelligibility index.
FIG. 2 is a diagram illustrating components necessary for calculating a speech intelligibility index and a speech intelligibility index of the present invention.
3 is a diagram illustrating a transfer function showing a correlation between a speech intelligibility index and speech intelligibility according to an embodiment of the present invention.
4A and 4B are diagrams illustrating graphs of speech dynamic ranges in Korean according to speech types according to an embodiment of the present invention.
5A and 5B are diagrams illustrating a dynamic range for each frequency calculated according to an embodiment of the present invention.
6A and 6B are diagrams illustrating a method of calculating a speech intelligibility index to which Korean-based frequency importance is applied according to an embodiment of the present invention.
7A to 7C are diagrams illustrating a result of applying Korean-based frequency importance according to an embodiment of the present invention.
8 is a block diagram showing the structure of a hearing test apparatus according to an embodiment of the present invention.
9 is a block diagram showing the configuration of a speech dynamic range calculating unit according to an embodiment of the present invention.
10 is a flowchart illustrating a process of calculating hearing test data in the hearing test device according to an embodiment of the present invention.
FIG. 11 is a flow chart illustrating calculation of speech energy for each frequency that may be performed in a step before operation 101 of FIG. 10.

The present invention can be applied to various changes and can have various embodiments, and specific embodiments will be illustrated in the drawings and described in detail.

However, this is not intended to limit the present invention to specific embodiments, and should be understood to include all modifications, equivalents, and substitutes included in the spirit and scope of the present invention. In describing each drawing, similar reference numerals are used for similar components.

When an element is said to be 'connected' or 'connected' to another component, it is understood that other components may be directly connected to or connected to the other component, but other components may exist in the middle. It should be. On the other hand, when a component is said to be 'directly connected' or 'directly connected' to another component, it should be understood that no other component exists in the middle.

The terms used in the present application are only used to describe specific embodiments, and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this application, terms such as 'include' or 'have' are intended to designate the existence of features, numbers, steps, operations, components, parts or combinations thereof described in the specification, and one or more other features. It should be understood that the existence or addition possibilities of fields or numbers, steps, operations, components, parts or combinations thereof are not excluded in advance.

Hereinafter, a basic concept of a method of calculating a speech intelligibility index will be described with reference to the drawings in order to describe a specific embodiment of the present invention in more detail.

First, FIG. 2 is a diagram illustrating components necessary for calculating a speech intelligibility index and a speech intelligibility index of the present invention.

)의 값이 500Hz와 1000Hz에서 1.0이고 2000Hz에서는 0.5이며, 4000Hz 및 5000Hz에서는 0으로 표시되고 있고, BIF(

)값이 BAF가 산출된 주파수 대역마다 표시되고 있다. 그리고 주파수 대역별로 BAF와 BIF를 곱한 값들과, 상기 주파수 대역별로 BAF와 BIF를 곱한 값들을 더한 값이 더불어 표시되고 있다. 이 때 주파수 대역별로 BAF와 BIF를 곱한 값들을 전체 더한 값이 SII(어음명료지수, Speech Intelligibility Index)이다. 즉, 어음명료지수는 하기 수학식 1과 같이 산출될 수 있다.Looking at the bottom of the graph of Figure 2 first BAF (

) Is 1.0 at 500 Hz and 1000 Hz, 0.5 at 2000 Hz, and 0 at 4000 Hz and 5000 Hz, and BIF (

) The value is displayed for each frequency band where BAF is calculated. In addition, BAF and BIF multiplied by frequency band and BAF and BIF multiplied by frequency band are added together. At this time, the total value obtained by multiplying BAF and BIF by frequency band is SII (Speech Intelligibility Index). That is, the speech intelligibility index can be calculated as in Equation 1 below.

[Equation 1]

는 주파수대역 k 에서의 대역가청함수(band-audibility function, BAF)를 의미하며, 상기

는 주파수대역 k 에서의 주파수중요함수(band-importance function, BIF)를 의미한다. The SII is a speech intelligibility index (which can be expressed in a range between 0 and 1), and K is the total number of frequency bands, and the

Denotes a band-audibility function (BAF) in the frequency band k, wherein

Denotes a band-importance function (BIF) in the frequency band k.

The hearing test apparatus according to an embodiment of the present invention may calculate a speech intelligibility index as shown in Equation 1 above, and for this, may calculate a major element required for calculating the speech intelligibility index.

The main elements constituting the speech intelligibility index are 'bandwidth audible function' and 'frequency important function'. First, the 'bandwidth audible function' will be described as follows.

Referring to the graph portion of FIG. 2, a dynamic range of speech is set in the range of 20dB to 50dB in a range of 30dB, and the same configuration is shown for all frequency bands. In addition, the audible threshold value of the listener is indicated by a dotted line, the audible threshold of the listener is 10 dB between 500 Hz and 1000 Hz, and increased to 35 dB, which is the middle of the dynamic range of speech at 2000 Hz, as the frequency increases. Values continue to increase and are shown to fall outside the dynamic range of speech after 4000 Hz. BAF stands for Band Audibility Function, and the band audibility function can represent a value for the ratio of audible speech energy of the listener to the entire speech dynamic range. Specifically, since the listener can hear a dB sound above the audible threshold value, a listener's audible threshold value of 35 dB at 2000 Hz (corresponding to the middle of the entire speech dynamic range (20 to 50 dB)) is about 50%. It means that you can hear, and above 4000Hz, the listener's audible threshold is outside the dynamic range, meaning there is no audible sound. On the other hand, between 500Hz and 1000Hz, the audible threshold of the listener is displayed as 10dB, because the speech energy value of the entire dynamic range of speech (from the maximum level to the minimum level of speech energy) has a dB value greater than the audible threshold. This means that the user can hear all sounds in the corresponding frequency band.

The ratio of speech energy that can be heard by a listener (hearing test subject) in a dynamic range of speech (DR) is referred to as a band audible function, and the band audible threshold is measured by a hearing test apparatus according to an embodiment of the present invention. It can be calculated according to Equation 2 below.

[Equation 2]

는 주파수대역 i 에서의 신호대잡음비(signal-to-noise ratio, SNR)를 데시벨(dB) 단위로 표현한 값을 의미하고, 상기 K 는 어음강도의 최대값에서 어음의 실효치 레벨(speech root-mean-square level)을 뺀 값을 의미하며, 상기 D 는 어음의 역동범위를 의미할 수 있다. At this time, remind

Denotes a signal-to-noise ratio (SNR) expressed in frequency band i in decibels (dB), and K is a speech root-mean- at the maximum value of speech intensity. square level), and D may mean a dynamic range of speech.

In the case of the conventional English version of the band audible function, it is standardized to ㅁ 15 dB based on the effective level of speech, and this range was adopted based on the recognition of speech measured in various SNR situations and the dynamic range of standardized speech. . That is, in the case of English, the dynamic range of speech is 30 dB over all frequencies, and the value of the band audible threshold is determined between 0 and 1. 0 means that the listener's hearing threshold is outside the dynamic range of speech and no speech can be heard, and 1 means that the listener's hearing threshold is better than the speech threshold in all frequency bands, so that all speech can be heard. Can be.

 The second component constituting the speech intelligibility index is a frequency critical function (BIF), and the hearing test apparatus according to an embodiment of the present invention may select or calculate a frequency important function for a test recording file. The frequency critical function refers to an index indicating the degree of importance of speech recognition in comparison with other frequency bands. That is, the importance of each frequency band is calculated as a relative importance by fixing the total importance to 1, and adding up the importance in each frequency band results in a sum of 1. For example, in the case of the frequency important function calculated based on the English sentence list, the 1,370 Hz (CF; center frequency) band is 0.1099, the 150 Hz (center frequency) band is 0.0315, and the 2,900 Hz (center frequency) band is calculated as 0.0646. Can be. As described above, the frequency importance may have different values for each frequency band, and according to the above example, it can be seen that the most important frequency band for speech recognition is a 1370 Hz band having a value of 0.1099 (10.99%). And if you add up the importance of each frequency band, the sum can be 1 (100%).

The hearing test apparatus according to an embodiment of the present invention may obtain frequency importance using a multi-step procedure and an iterative approach. The first step is to divide the various noise conditions and various frequency bands using SNR using filtering, and then measure speech perception in various conditions. According to an embodiment of the present invention, by using a monosyllabic table, SNR is 10 in total at 2 dB intervals from -10 dB to +8 dB, and filtering can measure the speech perception in a total of 13 frequency bands to obtain the frequency critical function. .

In the second step, the hearing test apparatus may obtain a rough correlation between the average speech recognition and speech intelligibility index values under each condition. In general, maps of speech can be displayed by changing from percentage (0 to 100%) to exponent (between 0 and 1). At this time, the rough relationship between the two indicators is formed through the graph. To express the graph smoothly, the following four laws can be followed. (1) One SNR graph should not overlap another SNR graph. (2) In the same SNR condition, there is only one intersection between a low-pass filter and a high-pass filter. (3) As SNR increases, speech recognition increases. (4) Under the same SNR condition, the two graphs (low-pass & high-pass filters) should end up with a similar score as much as possible.

The graph for the transfer function may be accompanied by a process by graphing (eg, cubic spline curve fitting method) in order to be calculated in the form of a smooth curve.

The third step is to find the speech recognition index and the corresponding speech intelligibility index based on the graphs calculated in the second step. For example, the intersection of the two curves (low-pass & high-pass filters) calculated from the highest SNR is 50% of the total speech map when the map is 100 with the most favorable condition for speech recognition. Since it can be seen as a point, it can be regarded as a value corresponding to 0.5 SII. The apparatus according to the embodiment of the present invention can find dozens of corresponding points in this way and express them in a graph. In addition, the following Equation 3 may be used to calculate a mathematical formula that can best estimate the inter-values of the corresponding graph.

[Equation 3]

In Equation 3, S is speech, P is the proficiency of the speaker and listener's speech and listening skills (usually fixed to 1), A is the speech intelligibility index value, and Q and N are constants for representing the curve well. It means the fitting constant.

The fourth step is to use the inverse equation in Equation 3 to replace the speech intelligibility index value (A) with the speech intelligibility map (S) so that the same equation holds, and then calculate all the average speech maps obtained in the first step. The inverse equation of equation 3 is used to find the corresponding speech intelligibility index. Then, all the average speech recognition maps are replaced with the corresponding speech intelligibility index values, which appear as cumulative distributions that include the speech intelligibility index values in the previous frequency band rather than the values corresponding to each frequency band. In order to change this to the speech intelligibility index value for each frequency band, the value of the frequency band one level below is subtracted from the highest frequency band value. In this way, the speech intelligibility index value for each frequency band can be obtained.

The fifth step is to average the values of the low-pass filter and the high-pass filter for the same frequency band in the same SNR. However, if one value is 0 or less than 0, it can be obtained by using only the other value that is positive. Then, it is possible to average the values in all SNR conditions in the same frequency band. Then, there is one integrated average value for each frequency band, and this is a cumulative distribution. If the value of the widest frequency range is not 1, it can be fixed to 1, and the remaining values can be expressed as a relative ratio using a ratio formula. In addition, if the cumulative distribution value whose maximum value is changed to 1 is replaced with each frequency value, the frequency importance for each final frequency band can be obtained.

3 is a diagram illustrating a transfer function showing a correlation between a speech intelligibility index and speech intelligibility according to an embodiment of the present invention.

The hearing test apparatus according to the embodiment of the present invention may convert the speech intelligibility index calculated according to the transfer function of FIG. 3 shown above into speech recognition.

3 shows a graph of a transfer function produced with an arbitrary value. In the graph, the X axis is the speech intelligibility index (SII) value, and the Y axis is the speech map value. The transfer function graph shows the correlation between speech intelligibility index and speech recognition. In addition, the transfer function may be used to calculate an approximate value of speech recognition through a speech intelligibility index (calculated based on a frequency important function and a band audible function).

For example, if the speech intelligibility index obtained through the listener's hearing threshold, frequency critical function, and band audible function was 0.2 SII before wearing the hearing aid, it was found that when applying the transfer function of FIG. 3, speech recognition of about 25% was expected. Can be. Also, if the speech intelligibility index score re-measured after wearing the hearing aid was 0.5 SII, it can be seen that speech recognition would be calculated to be about 85%. Accordingly, it can be seen that the change in speech perception before and after wearing the hearing aid is about 60%, from 25% to 85%. Such a speech recognition calculation method can be used to predict the effect of wearing a hearing aid by applying it to a child or an elderly person having difficulty in measuring actual speech recognition.

Conventionally, frequency importance, which is mainly suited to the characteristics of English, has been applied to various data required for hearing test including speech intelligibility index. However, it was found that the importance of frequency according to the frequency band differs according to the language, gender, and type of speech (single syllable, sentence, discourse). That is, the frequency bands considered important in listening and recognizing Korean and English sentences may be different. However, the method that has been applied in the related art has problems such as applying the frequency importance calculated based on the English speech even when the test is performed using Korean without excluding differences in the languages used. Accordingly, various embodiments of the present invention can calculate the frequency importance based on Korean characteristics and apply it to calculate hearing test data.

Hereinafter, a process for calculating auditory test data according to Korean characteristics according to an embodiment of the present invention will be described.

4A and 4B are diagrams illustrating graphs of speech dynamic ranges in Korean according to speech types according to an embodiment of the present invention.

In addition, FIG. 4A shows a speech dynamic range of a male, and FIG. 4B shows a graph of the speech dynamic range of a female.

Specifically, the speech dynamic range of Korean according to the type of speech may be measured as follows. The syllable dynamic range of single syllables ranged from 7.79 dB (CF: 150 Hz) to 27.71 dB (CF: 1,170 Hz) for men and 29.73 dB (CF: 1,370 Hz) to 10.61 dB (CF: 150 Hz) for women. . The speech dynamic range of sentences ranges from 8.96 dB (CF: 150 Hz) to 23.83 dB (CF: 7,000 Hz) for men and 11.06 dB (CF: 150 Hz) to 23.19 dB (CF: 5,800 Hz) for women. The speech dynamic range of the speech was 8.64 dB (CF: 150 Hz) for males and 24.66 dB (CF: 7,000 Hz) for males, and 23.52 dB (CF: 8,500 Hz) for females at 11.61 dB (CF: 250 Hz).

The syllable dynamic range of single syllables was analyzed to be significantly wider than the dynamic range of sentences and discourses in the middle frequency band for both men and women. For men, when the central frequency was 350 ~ 2900Hz, the dynamic range of single syllables was significantly greater than the dynamic range of sentences and discourses. It was wide.

In this frequency band, the difference in speech dynamic range between single syllables and sentences was between 1.36 dB (CF: 350 Hz) and 7.16 dB (CF: 1,370 Hz), and the difference in speech dynamic range between single syllables and discourse was 2.4 dB (CF: 350 Hz). At 6.95 dB (CF: 1,170 Hz). In the case of women, when the center frequency was 450 ~ 3,400 Hz, the dynamic range of monosyllables was significantly wider than that of sentences and discourse ( p <0.05). In this frequency band, the difference in speech dynamic range between monosyllables and sentences was between 2.23 dB (CF: 570 Hz) and 7.6 dB (CF: 1,370 Hz), and the difference in speech dynamic range between monosyllables and discourse was 2.32 dB (CF: 3,400 Hz). It was between 7.43 dB (CF: 1,370 Hz). 4A and 4B, the dynamic range of the sentence and the discourse was not significantly different. In the case of men, there was no significant difference in the rest of the frequency bands except for the center frequencies of 350 and 450 Hz, and in the case of women, there was no significant difference except when the center frequencies were 1,600 and 1,850 Hz.

5A and 5B are diagrams illustrating a dynamic range for each frequency calculated according to an embodiment of the present invention.

FIG. 5A shows the speech dynamic range of men (the value between the maximum and minimum levels of speech energy), and FIG. 5B shows the speech dynamic range of women. In the graph, the solid line shows the maximum and minimum values of speech energy for single syllables, the bold dotted line shows the maximum and minimum values of speech energy for the discourse.

5A and 5B show the distribution of speech dynamic ranges L99-L30 through the maximum and minimum levels of each speech type. As can be seen in FIGS. 5A and 5B, it can be seen that, in both men and women, monosyllables have a distribution of a larger maximum level (L99) and a smaller minimum level (L30) than sentences and discourses.

According to the study, the speech showed a speech dynamic range of about 8 to 30 dB in the case of single syllables, a speech dynamic range of about 9 to 23 dB in the case of sentences, and 9 to 25 dB in the case of speech. That is, in the comparison of the speech dynamic range according to the speech, the speech dynamic range of a single syllable was found to be wider in the middle frequency range than the speech dynamic range of a sentence or discourse. On the other hand, no significant difference was found in the speech dynamic range of sentences and discourse in most frequency bands.

Also, in the Korean speech dynamic range, significant differences were found for each frequency according to gender. For women, speech dynamic range of about 10 ~ 30dB was calculated according to frequency, and for male, speech dynamic range of about 7 ~ 27dB Was calculated. That is, it was found that the speech dynamic range of women was wider in most frequency domains than men.

Accordingly, in the hearing test method according to an embodiment of the present invention, a result may be calculated by applying a data value (eg, frequency importance) in consideration of the difference in speech dynamic range according to the type of gender and stimulation sound as described above.

Specifically, the hearing test apparatus according to an embodiment of the present invention may reflect a difference in the speech dynamic range according to the type of speech and the speech dynamic range according to gender. For example, the hearing test apparatus according to an embodiment of the present invention applies a speech dynamic range for each frequency corresponding to a corresponding attribute according to the gender of the voice and the type of speech (which corresponds to a single syllable, a sentence, or a discourse) of a test recording file can do.

6A and 6B are diagrams illustrating a method of calculating a speech intelligibility index to which Korean-based frequency importance is applied according to an embodiment of the present invention.

As the frequency importance, each importance value for 21 frequency bands can be applied as shown in the figure, and the sum of frequency importance of the entire band is 100%. At this time, different values of speech energy and frequency importance may be used depending on the gender (male or female) of the person speaking, and different speech energy and frequency importance may be applied depending on the level of stimulus (talk, sentence, monosyllable). .

Referring to FIG. 6B, a graph in which a hearing threshold of a listener (audience test subject) is displayed for each frequency and connected by a line is shown together with a speech energy graph. The ratio between the maximum / minimum level of the audible range that a listener can hear in each frequency band (the ratio between the maximum / minimum speech energy of the listener in a specific frequency band) Is the band audible value (follows the band audible function BAF). In FIG. 6B, a band audible value of a listener is calculated and displayed for each frequency band (21 bands). The hearing test apparatus according to an embodiment of the present invention can calculate the SII for each frequency band by multiplying the frequency importance (I) calculated for each frequency band and the band audible value (A), and adding SII for each band to all bands total SII (Speech Clearance Index) will be calculated.

Referring to the numerical values shown in FIG. 6B, assuming that the frequency is 1 from the low frequency, the hearing threshold has a smaller value than the speech energy up to the frequency 10, so that all signals can be sufficiently heard. From the frequency band where the arrow is displayed, the hearing threshold starts to have a value greater than the speech energy, so it can be seen that the listener is unable to hear the sound gradually.

For frequency band 11, the BAF (A) value is 0.9 and the BIF (I) value is 4.4. This means that the listener can hear 90% of the total speech energy in the frequency band, and the importance of the frequency is 4.4% (0.044). In addition, SII for each frequency band may be calculated as 3.96 (%), which is a result of multiplying the two values (0.9 * 0.044 = 0.0396). In addition, if all the calculated SII values for each frequency are added, a total SII value can be calculated. In FIG. 6B, it can be seen that the value is calculated as 0.6653. This may mean that about 66.53% of the audible test data can be audible.

As shown in the figure, the hearing test apparatus according to an embodiment of the present invention calculates SII for each frequency band by multiplying the frequency importance calculated for each frequency band and the band audible value, and then calculating the SII for each frequency band. Can be added to calculate the final SII.

7A to 7C are diagrams illustrating a result of applying Korean-based frequency importance according to an embodiment of the present invention.

FIG. 7A shows that the line facing downward indicates the subject's audible threshold, and the two lines across the entire frequency band show the maximum / minimum value of speech energy. Assuming that the test speech data (Korean and English) have the same speech dynamic range, the difference in the value of SII may be noticeable in Korean and English according to the frequency importance value for each frequency band. Referring to FIG. 7B, it can be seen that BIF (frequency importance) is set differently for English and Korean, with the band audible value according to the BAF (band audible function) being the same for each frequency band. Accordingly, the ultimately derived SII value is 0.49 for English and 0.59 for Korean.

Referring to FIG. 7C, it is possible to more clearly guess the difference between SII values for English and Korean. When the English SII value (0.49) is converted into a speech recognition value according to the transfer function, this can be calculated as 85%. However, in the case of Korean, if the SII value (0.59) is converted into a speech recognition value according to the transfer function, this can be calculated as 96%. As described above, in the hearing test for speech having the same speech dynamic range, it can be seen that the difference between the final speech intelligibility index and speech recognition value occurs depending on which language the speech to be tested is in.

The reason why the frequency importance is calculated differently for each language is because the attributes of the syllable structure and intonation of the language are different, and the details are as follows.

The dynamic range of Korean has a narrow dynamic range compared to other languages (English, Cantonese) at low frequencies below 455Hz based on the center frequency, and the dynamic range of English is the narrowest in the medium frequency range from 455Hz to 4050Hz, and more than 4050Hz In the high frequency band, the dynamic range of Korean and Cantonese was similar, but the dynamic range of English has a relatively wide dynamic range compared to other languages. In other words, the dynamic range of English is relatively narrow in the middle frequency band and wide in the high frequency band compared to Korean and Korean is opposite to English.

Also, in the structure of syllables, Korean is different from other languages. For example, the syllable structure of the English language can form syllables as one vowel, such as / a / in the shortest case, but in syllables such as / strength /, the number of consonants is significantly higher than that of vowels. On the other hand, Korean has a structure in which the number of consonants and vowels remains similar to that of English, such as / a /, / angle /, and angle /. This may suggest that Korean, which has a lot of vowel use relative to English, may use low-frequency energy more, and this may imply that Korean has a greater effect on the low-frequency region of speech than English.

Accordingly, the hearing test apparatus according to the embodiment of the present invention can calculate the speech intelligibility index by applying the frequency importance according to the characteristics of the Korean language. For example, when it is confirmed that the language of the test recording file is Korean, the hearing test apparatus according to an embodiment of the present invention may be controlled to apply the frequency importance calculated according to Korean characteristics. For example, if the recording file for the test is Korean, the device according to the present invention has the frequency set as the default frequency in the low frequency band (assuming it is divided into three levels of low frequency, medium frequency, and high frequency). The importance of increasing a certain value (the first adjustment value) in the importance level is applied, and the importance of increasing the constant value (the second adjustment value) is applied in the middle frequency band, and the constant value (the third adjustment value) is applied in the high frequency band. It can be controlled to apply the reduced importance. At this time, the size of the decrement and the increment is set to be the same, so that the sum of the total frequency importance remains 1 (or 100%) even after the frequency importance adjustment.

8 is a block diagram showing the structure of a hearing test apparatus according to an embodiment of the present invention.

Referring to FIG. 8, an apparatus according to an embodiment of the present invention may include a storage unit 810, an input unit 820, a display unit 830, an audio processing unit 840, and a control unit 850. In addition, the control unit 850 may include a speech dynamic range calculation unit 851, a band audible calculation unit 852, a frequency importance calculation unit 853, a test file type confirmation unit 854, a frequency importance application unit 855, and a speech recognition capability determination unit 856. .

First, the storage unit 810 may store data for calculating a band audible function, a frequency critical function, and a speech intelligibility index. Also, the storage unit 810 may store a transfer function required to calculate speech recognition from a speech intelligibility index and data related thereto. In addition, the storage unit 810 may store data regarding frequency importance that is differently set according to the type of test file (voice type, language type, gender of voice). The frequency importance stored in the storage unit 810 may be a value calculated by the frequency importance calculation unit 853. Alternatively, the storage unit 810 may include information on frequency importance set as a default. In this case, the frequency importance that is basically set in the device may be a frequency importance value (frequency importance calculated based on English) conventionally used.

Also, the storage unit 810 may store data required to calculate a dynamic range of speech for each frequency band. In addition, the storage unit 810 may store various data required to determine speech recognition ability of the hearing test subject.

The display unit 820 may display information required for performing a hearing test in the hearing test apparatus according to an embodiment of the present invention and test result information to a user.

The input unit 830 may receive values required to perform a hearing test in the hearing test apparatus according to an embodiment of the present invention. For example, the input unit 830 inputs the hearing test subject's response to the test file (eg, when a hearing test is performed in a manner of typing a letter with respect to what is heard, a typed word or sentence may be applied). Can receive

The audio processing unit 840 may output a test file (a file obtained by recording a single syllable, a sentence, or a discourse in a user's voice to test the speech recognition ability of a test subject). In addition, according to various embodiments of the present disclosure, when performing a speech recognition test, the audio processing unit 840 may receive a sound spoken by a test subject when a test is performed in a manner in which speech to be heard is spoken out.

The control unit 850 may include a speech dynamic range calculation unit 851, a band audible calculation unit 852, a frequency importance calculation unit 853, a test file type confirmation unit 854, a frequency importance application unit 855, and a speech recognition ability determination unit 856.

First, the speech dynamic range calculating unit 851 will be described with reference to FIG. 9.

9 is a block diagram showing the configuration of a speech dynamic range calculating unit according to an embodiment of the present invention.

The speech dynamic range calculating unit 851 may include a silence removing unit 851a and a histogram level calculating unit 851b to calculate a speech dynamic range of a test file. When the test file is input to the hearing test apparatus according to an embodiment of the present invention, the silence removing unit 851a may control silence of the corresponding file. In this case, silence removal can be performed only for silence maintained for 250 ms or more. When the silence time is 200 ms or more, it is determined that it is outside the range of silence that occurs naturally when speech is produced, so the silence removal reference time can be set to 250 ms.

Subsequently, the histogram level calculator 851b may derive cumulative histogram levels through various steps on the data from which silence is removed. The histogram level calculator 851b may perform a process of deriving a cumulative histogram level of recorded data as follows. (1) After calculating the root mean square level of the recording data, each recording file is normalized to 65 dB SPL. (2) The signal envelope is calculated in 21 critical bands with a center frequency ranging from 150 to 8,600 Hz. For reference, these 21 frequency band ranges may be based on the threshold band range when calculating the speech recognition index. (3) Curve the signal envelope smoothly through a raised cosine window calculated in units of 1,000ths of a second using a mat trap. (4) The curved signal envelope is subsampled with 50% overlap of the window length and converted into dB SPL. (5) Change the cumulative signal envelope distribution level to a decibel signal envelope histogram to derive the dynamic range of speech.

For reference, the cumulative histogram level may show the distribution of the cumulative signal envelope of speech in the frequency band. For example, the 99% cumulative histogram level (L99) includes a peak level in the measured speech signal or It can point to 99% or less. The 30% cumulative histogram level L30 may include a minimum level of the measured speech signal or indicate a portion of 30% or more. In other words, the L99 refers to the maximum level of speech, and the L30 refers to the minimum level of speech. (The speech dynamic range represents the value obtained by subtracting the minimum level from the maximum level (L99 ?? L30).) As described above, the speech dynamic range calculator 851 calculates speech dynamic ranges for each frequency band from a recording file for testing. Can be.

The band audible calculator 852 may determine the maximum and minimum levels of speech energy based on the speech dynamic range calculated by the speech dynamic range calculator 851. In addition, the band audible calculator 852 may calculate a band audible value that follows a band audible function based on a speech dynamic range calculated for each frequency band. At this time, the band audible function follows Equation 2.

[Equation 2]

는 주파수대역 i 에서의 신호대잡음비(signal-to-noise ratio, SNR)를 데시벨(dB) 단위로 표현한 값을 의미하고, 상기 K 는 어음강도의 최대값에서 어음의 실효치 레벨(speech root-mean-square level)을 뺀 값을 의미하며, 상기 D 는 어음의 역동범위를 의미할 수 있다. At this time, remind

Denotes a signal-to-noise ratio (SNR) expressed in frequency band i in decibels (dB), and K is a speech root-mean- at the maximum value of speech intensity. square level), and D may mean a dynamic range of speech.

At this time, the audible threshold of the test subject required to calculate the band audible value may be calculated through an audible threshold measurement unit (not shown). For example, the audible threshold measurement unit may determine a user's audible threshold level based on a reaction record of a subject according to reproduction of a test recording file. According to various embodiments of the present disclosure, the audible threshold measurement unit may determine a user's audible threshold level based on test data on not only speech but also pure tone hearing.

The frequency importance calculator 853 may perform a function of calculating the frequency importance for the test file. The frequency importance calculator 853 may calculate the frequency importance based on the recognition rate of the test subject corresponding to the frequency band and the speech dynamic range of speech based on the test results already performed. In addition, the frequency importance calculator 853 may calculate frequency importance differently for each language type, gender type, and speech type recorded in the test file. For example, the frequency importance calculator may calculate a frequency importance level corresponding to a high frequency region of 4050 Hz or higher for an English sentence as a value 0.2 higher than the frequency importance in a Korean sentence. In this case, the sum of the frequency importance calculated by the frequency importance calculator 853 for the entire band may be set to 1 (or 100%).

In addition, according to various embodiments, the frequency importance calculator 853 may calculate the frequency importance as an increment or a decrease of a predetermined basic frequency importance (eg, frequency importance calculated based on English).

The test file type checking unit 854 may check the type of the test file. Specifically, the test file type confirming unit 854 may confirm whether the gender of the voice recording the test file is female or male. In addition, the test file type confirmation unit 854 may check whether the language type of the test file is Korean or a foreign language. In addition, the test file type checking unit 854 may determine whether the speech type is a discourse, a sentence, or a single syllable. When the test file is input to the device, the test file type checking unit 854 simulates and reproduces the file itself to analyze the recorded signal information, and thus can determine the type of the test file. Alternatively, the test file type checking unit 854 may determine a speech type, a language type, and the like as a method of determining a predetermined identification value for the test file.

The frequency importance applying unit 855 may apply a corresponding frequency importance based on the identified type of the test file. For example, when it is confirmed that the test file is Korean, the frequency importance applying unit 855 sets the frequency importance, so that the frequency importance value corresponding to the Korean language calculated by the frequency importance calculating unit 853 and stored in the storage unit 810 is applied. Can be controlled.

The speech recognition ability determination unit 856 may calculate a speech intelligibility index and speech recognition of the hearing test subject. At this time, when calculating the speech intelligibility index, the speech recognition ability determining unit 856 may calculate the speech intelligibility index based on Equation 1 below.

 [Equation 1]

는 주파수대역 k 에서의 대역가청함수(band-audibility function, BAF)를 의미하며, 상기

는 주파수대역 k 에서의 주파수중요함수(band-importance function, BIF)를 의미한다. The SII is a speech intelligibility index (which can be expressed in a range between 0 and 1), and K is the total number of frequency bands, and the

Denotes a band-audibility function (BAF) in the frequency band k, wherein

Denotes a band-importance function (BIF) in the frequency band k.

The speech recognition capability determining unit 856 may calculate speech recognition based on the calculated speech intelligibility index. The speech recognition capability determining unit 856 may apply a transfer function to the speech intelligibility index to calculate a rough estimate of speech recognition. The transfer function indicating the correlation between the speech intelligibility index and speech intelligibility was described in FIG. 3 and will be omitted.

10 is a flowchart illustrating a process of calculating hearing test data in the hearing test device according to an embodiment of the present invention.

FIG. 11 is a flow chart illustrating calculation of speech energy for each frequency that may be performed in a step before operation 101 of FIG. 10.

Referring to FIG. 10, the control unit 850 may perform operation 101 of calculating a band audible value, and then the control unit 850 may perform operation 103 of applying frequency importance. Thereafter, the controller 850 may perform operation 105 of calculating a speech intelligibility index based on the frequency importance and the band audible value.

The band audible value and the frequency importance can all be set corresponding to 21 frequency bands. When calculating the speech intelligibility index, the controller 850 first calculates the speech intelligibility index for each frequency by multiplying the band audible value corresponding to the same frequency by the frequency importance, and then adds the calculated speech intelligibility indexes to the final speech intelligibility. The index can be calculated. Thereafter, the controller 850 may perform operation 107 of calculating speech recognition based on the calculated speech intelligibility index. The process of calculating speech recognition may be omitted. Also, the operation of calculating speech recognition may be calculated using a transfer function that is a relational expression between the speech intelligibility index and speech recognition. Description of each data and transfer function required to calculate each speech intelligibility index has been described in the description of FIGS. 2 and 3 and will be omitted.

In the present invention, speech dynamic range (or speech energy), which is information required to calculate a band audible value corresponding to each of a plurality of bands, may be calculated for each frequency band (eg, 21 bands). The operation of calculating the speech dynamic range for each frequency band will be described with reference to FIG. 11.

Referring to FIG. 11, the controller 850 may perform operation 111 of checking the input of the test file. Thereafter, the controller 850 (eg, silence removing unit 851a) may perform operation 113 to remove silence from the inputted file. At this time, the control unit 850 may perform the removal only for silences that are maintained for 250 ms or more in order to discriminate from silences essential for general speech vocalization.

Thereafter, the controller 850 (eg, the histogram level calculator 851b) may perform operation 1153 of calculating signal envelopes in a plurality of threshold bands with the silence removed data. At this time, the threshold band may be set to 21 regions having a center frequency of 150 to 8600 Hz.

Thereafter, the controller 850 may perform operation 117 of curveting the signal envelope. Thereafter, the controller 850 may perform operation 119 to perform sub-sampling and unit conversion of the curved signal envelope.

Thereafter, the controller 850 may calculate speech dynamic ranges for 21 frequency bands by performing 121 operations of converting the cumulative signal envelope distribution level into a decibel signal envelope histogram.

Previously, an operation of measuring speech recognition ability through a hearing test apparatus according to an embodiment of the present invention was described. However, the present invention is not limited thereto, and according to various embodiments, an operation for calculating frequency importance may be performed based on statistics on speech recognition results.

Specifically, for example, the controller 850 may extract a user's speech intelligibility index for each frequency band that is pre-classified based on a result of speech recognition of test files of a plurality of hearing test subjects. The speech intelligibility index for each frequency band is calculated before the total SII (voice intelligibility index) is calculated, and thus can be extracted without a separate calculation process. At this time, the control unit 850 may select only the speech intelligibility indices for each frequency band of the hearing test subjects whose average result of the total speech intelligibility index is average (determined to be normal speech recognition ability). Thereafter, the controller 850 may classify speech intelligibility indices for the same test file. In addition, the control unit 850 may determine a frequency band having a high partial speech intelligibility index for the same test file as a frequency band having a high recognition rate and further weight the frequency importance of the corresponding band. For example, in the state where the existing frequency importance for the 1 band and the 2 band are set to 50% and 50%, respectively, the statistical results show that the speech indices of the hearing test subjects for the 1 band are higher than a certain value. When doing so, the control unit 850 may set the importance for one band to a high value (eg, 1%). In addition, since the values of the importance increment and the importance decrease must be the same, if the importance is increased in a specific band, the importance may be reduced by the increased importance in the remaining bands. That is, as the importance increases by 1% in the first band, the importance may decrease by 1% in the second band. If the importance increase is 20% in only one of the 21 bands, the remaining bands (20 bands) can be divided into the same increments, and the importance can be reduced. In this case, the frequencies in the remaining 20 bands The importance can be reduced by 1% each.

According to various embodiments of the present disclosure, as soon as a test file is input, the control unit 850 may perform an operation of analyzing the recording content (eg, frequency band) of the file to check the gender of the voice. Alternatively, when the test file is output and provided to the hearing test subject, the control unit 850 analyzes the frequency of the sound output and performs speech recognition, such as speech attributes (speech type (single syllable, sentence, discourse), gender, language type, etc. ). And accordingly, the control unit 850 (eg, frequency importance application unit 855) may apply the frequency importance corresponding to the analyzed test file type.

Although the present invention has been described in detail with reference to the above-mentioned examples, those skilled in the art can make modifications, alterations and modifications to the examples without departing from the scope of the present invention. In short, in order to achieve the intended effect of the present invention, it is not necessary to separately include all the functional blocks shown in the drawings or to follow all the order shown in the drawings in the order shown, and even if not, any number of technical aspects of the present invention described in the claims Note that it can fall within the scope.

810: storage unit
820: input unit
830: display unit
840: audio processing unit
850: control unit
851: Speech dynamic range calculator
852: Band audible calculator
853: frequency importance calculator
854: test file type confirmation unit
855: frequency importance section
856: speech recognition ability judgment unit

Claims (11)

In the hearing test method of the hearing test device,
Calculating a speech dynamic range for each frequency of a test file by a control unit of the hearing test apparatus;
Calculating, by the control unit, a band audible value that follows a band audible function for each frequency based on the calculated speech dynamic range;
Calculating, by the control unit, frequency importance for the same frequency from which the band audible value is calculated; And
Including, by the control unit, calculating a speech intelligibility index based on the band audible value and the frequency importance;
The operation of calculating the speech dynamic range for each frequency of the test file is
Removing silence in the data from the test file by the control unit;
Calculating, by the control unit, an average effective value level of the data from which the silence is removed, and then normalizing each recording file to a certain dB SPL;
Calculating, by the control unit, a signal envelope in a plurality of threshold bands;
Curving the signal envelope by the control unit;
Sub-sampling and unit-converting the curved signal envelope by the control unit; And
And calculating, by the control unit, a cumulative signal envelope distribution level into a decibel signal envelope histogram to calculate a dynamic range of speech. According to claim 1,
The operation of calculating the signal envelope is
Hearing test method based on the calculation of speech energy for each frequency, characterized in that the operation of calculating a signal envelope for 21 frequency bands having a center frequency of 150 to 8600 Hz with respect to the silenced data. According to claim 1,
The speech intelligibility index is calculated based on Equation 1 below,
[Equation 1]

The SII is a speech intelligibility index, the K refers to the total number of frequency bands, and the

Denotes a band-audibility function (BAF) in the frequency band k, wherein

Is a frequency-important function in the frequency band k (band-importance function, BIF). According to claim 1,
The operation of removing silence in the data is
A hearing inspection method based on calculating speech energy for each frequency, characterized in that it is an operation to remove the silence only when the duration of silence is 250 ms or more. According to claim 1,
The operation of curving the signal envelope is
This is an operation to curve the envelope through a raised cosine window, calculated in 1000ths of a second,
The operation of sub-sampling and converting the curved signal envelope
A method of hearing evaluation based on speech energy calculation for each frequency, characterized in that the curved signal envelope is sub-sampled by overlapping the window length by 50% and converted into dB SPL units. According to claim 1,
The band audible function
The ratio of speech energy that can be heard by a listener based on the listener's hearing threshold in a given frequency band, follows Equation 2,
[Equation 2]

remind

Denotes a signal-to-noise ratio (SNR) expressed in frequency band i in decibels (dB), and K is a speech root-mean- at the maximum value of speech intensity. square level), and D is a hearing test method based on calculating speech energy for each frequency, characterized in that it means a dynamic range of speech. According to claim 1,
The hearing test method based on the calculation of speech energy for each frequency,
The controller further includes an operation of calculating a speech recognition map by applying a transfer function to the speech intelligibility index, wherein the speech recognition map is calculated according to Equation 3 below,
[Equation 3]

The S is speech, the P is the proficiency of the speaker and listener's speech and listening ability, the A is the speech intelligibility index value, and the Q and N are the fitting energy calculations for each frequency, Based hearing test method. According to claim 1,
The operation for calculating the frequency importance is
A frequency-based speech energy calculation-based hearing test method characterized in that the frequency importance is calculated according to the type of speech recorded in the test file and the gender of the voice. . The method of claim 8,
The operation for calculating the frequency importance is
As the test file is reproduced, determining the frequency band of the sound recorded in the test file to determine the gender and speech type of the voice, and applying frequency importance corresponding to the determined gender and speech type Hearing test method based on the calculation of speech energy by frequency, characterized in that. According to claim 1,
The operation for calculating the frequency importance is
And an operation of increasing and setting the frequency importance of the frequency band in which the speech intelligibility index value is higher than a preset value based on the speech intelligibility index for each frequency band of the hearing test subjects judged to have normal speech recognition ability Hearing test method based on calculation of speech energy by frequency. Calculate the speech dynamic range for each frequency for the test file, remove the silence in the data from the test file, calculate the average effective value level of the data from which the silence is removed, and normalize each recording file to a certain dB SPL Then, the signal envelope is calculated in a plurality of critical bands, the signal envelope is curved, the sampled curve is subsampled, and then the units are converted, and the cumulative signal envelope distribution level is converted into a decibel signal envelope histogram to reverse the speech. Speech dynamic range calculation unit for calculating a range;
A band assumption calculator for calculating a band audible value that follows a band audible function for each frequency based on the calculated speech dynamic range;
A frequency importance calculator for calculating frequency importance for the same frequency from which the band audible value is calculated; And
And a speech recognition capability determination unit that calculates a speech intelligibility index based on the band audible value and the frequency importance.

Images