Design and use method of digital triple tone test library based on Chinese characteristics

Abstract

The invention provides a design method based on a Chinese characteristic digital triple tone test library, which comprises the following steps: determining a more easily confused digital group through a similarity estimation formula and a perception distance formula, determining an audiometric material according to a preset condition, and correcting according to a correction factor formula to obtain a corrected audiometric material. The method comprises the following steps: randomly performing three-three combination to form a plurality of initial triple tones, performing scoring according to a preset scoring rule according to a preset speech frequency range to form a triple tone library conforming to frequency coverage, obtaining a to-be-optimized triple tone library according to a digital initial rhyme statistical table, determining a sum score of each triple tone in the to-be-optimized triple tone library according to an attribute table, and sorting and screening the to-be-optimized triple sound library according to the sum score to form an optimized triple sound library, traversing the optimized triple sound library, and screening out a final digital triple sound test library according to a preset balanced digital frequency in combination with the triple sound sum score. The use method comprises the following steps: disrupting the test library, adjusting stimulation sound intensity according to patient answers, calculating a signal-to-noise ratio, and obtaining a final hearing evaluation index.

Description

Technical Field

[0001] The present invention relates to the technical field of speech audiometry and hearing loss degree assessment, and in particular to a design of a digital triplet tone test library based on Chinese characteristics and a method for using the same. Background Art

[0002] Miller et al. (1951) found that among all speech materials, speech with digital content is the easiest to recognize, so they proposed to use numbers as test materials to explore the impact of context on speech intelligibility. In 2004, Smits et al. developed and verified the Dutch digital triplet test based on landline telephone for the first time under the research of Miller et al. (1951), and established the test process and test specifications of the digital triplet test (DTT). In the past decade, many countries including the United States, the United Kingdom, Australia, Germany, Poland, Switzerland and France have successively developed digital triplet tests based on landline telephones, the Internet and mobile phone APPs. In my country, Beijing Tongren Hospital affiliated to Capital Medical University took the lead in cooperating with the University of Southampton in the United Kingdom in 2012 to carry out research on recording Chinese digital triplet test materials and collecting normal values. In recent years, researchers engaged in the field of speech audiometry have also successively explored the application of DTT in patients with presbycusis and sensorineural hearing loss, and have also developed a number of DTT audiometry word lists such as the Ningbo dialect version.

[0003] The digital triple tone test method is a method to evaluate the hearing of the subject based on the speech recognition threshold. The design process and problems of the digital triple tone library are as follows:

[0004] (1) In the number selection step: Since the existing digital triplet test method was introduced from abroad, it continues the foreign number selection habits and selects numbers 0-9 that are pronounced as monosyllabic numbers in the language as materials, which does not conform to the pronunciation characteristics of Chinese numbers. It only uses the number of syllables as the standard, which is easy to confuse and mishear, thus affecting the accuracy of the test;

[0005] (2) In the step of combining to form a triplet library: it is only required that the frequency of each number appearing in each position of the triplet is as equal as possible, and there are no repeated numbers in each triplet, so as to optimize the combination of monosyllabic numbers into a triplet library. Since the digital acoustic frequency domain range is not considered when combining into the triplet library, the phonemes contained in the selected triplet numbers may not cover the main frequency range of speech sounds (500-4000Hz), and the triplet library does not consider the balance of vowels and tones. The difference in vowels and tones will cause listening errors and affect the accuracy of the test.

[0006] Due to the errors caused by the existing digital triple-tone design process, the test results of the digital triple-tone test method using the digital triple-tone library ultimately have large errors. Summary of the invention

[0007] The technical problem to be solved by the present invention is: since the prior art lacks consideration of the three aspects of confusion (hearing sensation), frequency coverage (acoustics) and balance of tones (phonology) of digital pronunciation, the triple sound library used for testing cannot cover the main frequency range (500-4000Hz) of speech sounds, and the difference in confusion and tones can also cause hearing errors, resulting in large errors in the established digital triple sound library, which ultimately reduces the credibility of the test results.

[0008] In order to solve the above technical problems, the technical solution of the present invention provides a design method of a digital triplet test library based on Chinese characteristics, comprising the following steps:

[0009] Obtain standardized digital audiometric materials, have multiple subjects conduct digital recognition experiments, and obtain the confusion matrix P based on the experimental results;

[0010] According to the similarity valuation formula, the similarity valuation p′ is calculated for the data in the confusion matrix P. ij , combined with the perceptual distance formula, calculate the perceptual distance d between numbers ij , determine the number groups that are more easily confused, and retain some of the numbers in the number groups according to the number of initials and finals in the pronunciation, and add corresponding numbers according to the number, continuity and tone of the eliminated numbers to determine the audiometric materials;

[0011] Conducting a homogeneity optimization test on the audiometric material, wherein the average speech recognition rate curve of each number in the audiometric material under different speech intensities or different signal-to-noise ratios is subjected to conformal interpolation, the speech recognition threshold of each number in the audiometric material before equalization and the average speech recognition threshold of all numbers in the audiometric material are determined, and the correction factor Factor is determined according to the correction factor formula, and the audiometric material is corrected according to the correction factor to obtain the corrected audiometric material;

[0012] The corrected audiometric materials are randomly combined into multiple initial triplets in groups of three, the frequencies of the initial consonant and final parts of each number in a single initial triplet are calculated and integrated to form a frequency range of each number, and according to a preset speech sound frequency range, initial triplets whose sum average of the frequency range of each number in each initial triplet is greater than or equal to the sum average of the upper and lower bounds of the preset speech sound frequency range are selected, and they are scored according to a preset scoring rule to form a triplet library that meets the frequency coverage;

[0013] A digital initial consonant, final consonant and tone statistical table is established according to each digital test, and triplets of the same initial consonant, the same final consonant and the same tone are removed from the triplet sound library that meets the frequency coverage, so as to obtain the triplet sound library to be optimized;

[0014] According to the frequency of use of initials, finals and tones, scores are assigned in descending order from high frequency to low frequency, and a table of dynamic frequency statistics of Mandarin initials, finals and tones and corresponding scores is generated; according to the scores of the triple sound library that meets the frequency coverage and the table of dynamic frequency statistics of Mandarin initials, finals and tones and corresponding scores, an attribute table with numbers, pinyin, phonetic symbols, frequency, frequency of use of initials, frequency of use of finals and frequency of use of tones is generated;

[0015] Determine the total score of each triplet in the triplet library to be optimized according to the attribute table, sort the triplet library to be optimized according to the total score, and select a preset number of triplets to form an optimized triplet library;

[0016] The optimized triplet tone library is traversed, and the final digital triplet tone test library is screened out according to the preset balanced digital frequency and the total score of the triplet tones.

[0017] Preferably, the similarity estimation formula is as follows:

[0018]

[0019] Among them, S ij It represents the probability of hearing the sound i as the sound j.

[0020] Preferably, the perception distance formula is as follows:

[0021] d ij =-ln(S ij ).

[0022] Preferably, the correction factor formula is as follows:

[0023]

[0024] Among them, SRT i represents the speech recognition threshold before equalization of each digit in the audiometric material, indicating The average speech recognition threshold for all digits in the audiometric material.

[0025] Preferably, determining the total score of each triplet in the triplet library to be optimized according to the attribute table comprises the following steps:

[0026] Determine the frequency, initial consonant usage frequency, final consonant usage frequency and tone usage frequency of each number according to the attribute table, and calculate the number of non-repeated initial consonants s of each triplet in the triplet library to be optimized n 、Number of finals sf The number of harmonic tones n , get the seven-dimensional attribute value of each triplet;

[0027] The seven-dimensional attribute values ​​are normalized to their maximum values ​​and the sum is calculated to obtain the total score.

[0028] Preferably, the preset speech sound frequency range is 500-4000 Hz.

[0029] Preferably, the preset number is 23 groups.

[0030] Preferably, the preset balanced digital frequency is greater than or equal to 6 and less than or equal to 8.

[0031] The technical solution of the present invention also provides a method for using a digital triplet test library based on Chinese characteristics, wherein the final digital triplet test library established by the design method of the digital triplet test library based on Chinese characteristics as described above comprises the following steps:

[0032] The order of triplets in the final digital triplet test library is disrupted, and the order of the numbers in the triplets is disrupted to obtain multiple groups of triplets in a disrupted order;

[0033] During the test, the patient was first tested based on the first set of triplets. The judgment standard was that when the patient gave a correct answer to all three numbers in the triplets, the answer was considered correct.

[0034] If the answer is correct, the intensity of the stimulus sound is reduced by a preset value, and the second set of triplet sounds is continued for testing; if the answer is incorrect, the intensity of the stimulus sound is increased by a preset value, and the second set of triplet sounds is continued for testing, until all sets of triplet sounds are tested;

[0035] The signal-to-noise ratio is calculated according to the fluctuation of the test intensity to obtain the final hearing evaluation index.

[0036] The technical solution of the present invention proposes a design method for a digital triplet test library based on the characteristics of Chinese. By calculating the similarity estimation between numbers and their corresponding perceptual distances, a number group that is more easily confused is determined. According to the number of initials and finals in the pronunciation, some numbers in the number group are retained. According to the number of eliminated numbers, continuity and tone, corresponding numbers are supplemented to determine the audiometric material.

[0037] Next, a homogeneity optimization test is performed on the audiometric material. By performing conformal interpolation on the average speech recognition rate curves of each number in the audiometric material under different speech intensities or different signal-to-noise ratios, the speech recognition threshold of each number in the audiometric material before equalization and the average speech recognition threshold of all numbers in the audiometric material are determined. The correction factor is determined according to the correction factor formula to correct the audiometric material and equalize the speech material.

[0038] Then, the corrected audiometric materials are randomly combined in groups of three to form multiple initial triplets, and the frequencies of the initial and final parts of each number in a single initial triplet are calculated and weighted integration is performed to form the frequency range of each number. According to the preset speech sound frequency range, the initial triplets whose sum average of the frequency range of each number in each initial triplet is greater than or equal to the sum average of the upper and lower bounds of the preset speech sound frequency range are selected, and the scores are performed according to the preset scoring rules to form a triplet library that meets the frequency coverage. A digital initial, final and tone statistics table is established based on the initials, finals and tones tested for each number, and triplets that test the same initial, final and tone are removed from the triplet library that meets the frequency coverage to obtain the triplet library to be optimized. The scores are assigned in descending order according to the frequency of use of initials, finals and tones from high to low frequency, and the dynamic frequency statistics of Mandarin initials, finals and tones and the corresponding score table are generated. An attribute table with numbers, pinyin, phonetic symbols, frequency, initial consonant frequency, final consonant frequency and tone is generated according to the scores of the triplet sound library that meets the frequency coverage requirements and the dynamic usage frequency statistics of Mandarin consonants, consonants and tones and the corresponding score table; the total score of each triplet in the triplet sound library to be optimized is determined according to the attribute table; the triplet sound library to be optimized is sorted according to the total score; and a preset number of triplets are selected to form the optimized triplet sound library.

[0039] Finally, the optimized triplet tone library is traversed, and the final digital triplet tone test library is screened out according to the preset balanced digital frequency and the total score of the triplet tones.

[0040] This technical solution solves the problem that the existing technology lacks consideration of the three aspects of confusion (auditory perception), frequency coverage (acoustics) and balance of tones (phonology) of digital pronunciation, resulting in the triple sound library used for testing failing to cover the main frequency range of speech sounds (500-4000Hz), and the differences in confusion and tones can also cause auditory errors, resulting in large errors in the established digital triple sound library, which ultimately reduces the credibility of the test results. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 An indication of the perceived distance between numbers provided by an embodiment of the present invention;

[0042] Figure 2 A schematic diagram of an average speech recognition rate curve under different speech intensities and an average speech recognition rate curve under different signal-to-noise ratios for each number provided in an embodiment of the present invention;

[0043] Figure 3 A digital frequency statistical representation diagram provided for an embodiment of the present invention;

[0044] Figure 4 A schematic diagram of digital frequency statistics provided by an embodiment of the present invention;

[0045] Figure 5 A flow chart of the design and use of a digital triplet tone test library provided by an embodiment of the present invention;

[0046] Figure 6 A schematic diagram of a Chinese digital triplet tone testing system provided by an embodiment of the present invention;

[0047] Figure 7 A DTT test flow chart provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application equally.

[0049] The embodiment of the present invention provides a design method of a digital triplet test library based on Chinese characteristics, comprising the following steps:

[0050] Step (1), preliminary material selection specific steps are as follows:

[0051] Step (1.1), recording pronunciation: select 6 speakers (3 men and 3 women) with Mandarin level 2A or above, and ask them to wear AKG C544 head-worn microphones to record the pronunciation of monosyllabic numbers 0-10 in Mandarin in a soundproof room. The audio is set to 22050Hz sampling rate and 16-bit sampling accuracy.

[0052] Step (1.2), screening representative pronunciation: 10 phonetics professionals score the digital pronunciation recorded in step (1.1). The scoring rule is to score on a scale of 1-5 based on five dimensions: whether the speaking speed is moderate, whether the tone is natural, whether the voice is clear, whether the sound quality is good, and whether the pronunciation (including the three aspects of finals, consonants and tones) is accurate. After scoring, the scores of the five dimensions are summed up and sorted from high to low according to the total score. Finally, the one with the highest score is screened out as the representative pronunciation.

[0053] Step (1.3), through the number recognition experiment, determine the audiometric materials: 0 and 2-10. According to multidimensional scaling (MIDS), calculate the auditory similarity of the pronunciation of the 11 numbers 0-10 in the representative pronunciation. The specific steps are as follows:

[0054] Step (1.3.1), 24 subjects with no history of ear disease and normal otoscopy and audiology examinations were selected to complete the digital recognition experiment in a soundproof room in a binaural state. The specific method is to play the recorded single digital pronunciation in noise and dynamically adjust the signal-to-noise ratio, and ask the subjects to choose the number they hear from the 11 numbers 0-10.

[0055] Step (1.3.2), based on the number recognition experiment, obtain the confusion matrix P of Chinese numbers 0-10;

[0056] Step (1.3.3) Calculate the similarity estimate p' ij :S ij Represents the probability of hearing the i-th consonant as the j-th consonant:

[0057]

[0058] Step (1.3.4) calculates the perceived distance d ij : Based on Shepard's law, calculate the negative value of the natural logarithm of the similarity:

[0059] d ij =-ln(S ij )

[0060] The perceived distance between numbers is Figure 1 shown.

[0061] according to Figure 1 As a result, the most easily confused number groups are: 1 and 7. The perceptual distance between 1 and 7 is the smallest, and they are most likely to cause confusion. This will cause the test subject to easily make the wrong choice even if they hear the sound because 1 and 7 are very similar, thus affecting the final test results. Therefore, 1 is removed (7 is not removed because the pronunciation of 7 contains one more initial consonant q than 1, that is, according to the number of initial consonants and finals in the pronunciation, some numbers in the number group are retained). Among 0-9, only 0 has the second tone, which may provide additional clues for the identification of 0. It is necessary to add 10, which is also the second tone (according to the number of eliminated numbers, continuity and tone, the corresponding numbers are supplemented to determine the audiometric material). Finally, the audiometric material consists of the pronunciation of 0 and the 10 numbers from 2 to 10, which is used for the material preprocessing of step (2), that is, the homogeneity optimization test.

[0062] Step (2), material preprocessing - homogeneity optimization test: "Homogeneity" is a basic indicator to ensure that the speech audiometry material has good reliability and sensitivity. In order to achieve the best test efficiency, the recognition difficulty of each of the 10 numbers 0 and 2-10 should be as homogeneous as possible, that is, to ensure that each number has the same speech clarity. The specific steps to achieve homogeneity optimization are as follows:

[0063] Step (2.1), masking noise production: the audiometric material obtained in step (1) is randomly superimposed 30 times to finally form a long-term average spectral noise with the same duration as the test tone;

[0064] Step (2.2), perception experiment: 20 subjects with no history of ear disease and normal otoscopy and audiological examinations were selected, and the 10 recorded digit pronunciations were played under the long-term average spectral noise obtained in step (2.1) and the signal-to-noise ratio was dynamically adjusted. The subjects were asked to choose the digits they heard, and whether each digit was correctly recognized at each signal-to-noise ratio was recorded;

[0065] Step (2.3), test results: The average speech recognition rate of each digit under different speech levels or different signal-to-noise ratios (SNR) is obtained by averaging the original data obtained from people with normal hearing, and a curve is drawn as follows: Figure 2 As shown. The pchip function of Matlab software was used to perform shape-preserving interpolation on the curve, and the speech intensity or signal-to-noise ratio when the accuracy rate was 50% was obtained as the speech recognition threshold (SRT) before equalization of each number. The speech recognition thresholds of the numbers 0 and 2-10 in quiet were 8.4, 3.9, 3.6, 5.0, 5.0, 3.2, 5.0, 3.9, 1.9, and 8.5 dB A, respectively. The speech recognition thresholds of the numbers in noise were -13.1, -16.9, -14.0, -16.7, -11.5, -12.8, -15.5, -17.7, -15.0, and -11.4 dB SNR, respectively.

[0066] Step (2.4), optimization scheme: By setting the speech recognition threshold (SRT) of the number i ) minus the average speech recognition threshold of all digits Calculate the correction factor:

[0067]

[0068] The correction factors for each number in quiet are 3.5, -0.9, -1.2, 0.2, 0.2, -1.6, 0.2, -0.9, -3.0, 3.7 dB, and the correction factors for each number in noise are 1.4, -2.4, 0.5, -2.3, 2.9, 1.7, -1.0, -3.2, -0.6, 3.1 dB. The correction factors are applied to the original audio sound intensity of the numbers to align the 50% correct recognition probability of all numbers, equalize the speech materials, and obtain the corrected audiometric materials.

[0069] Step (3), create a triplet sound library:

[0070] Smits and Houtgast found through multiple experiments that randomly generated three-digit numbers (triplets) are the most suitable test materials. This form can reduce the requirements for auditory memory and cognitive burden. At the same time, due to the high randomness, the learning effect is small, and the influence of rhythm and co-articulation is eliminated. At the same time, according to the previous research practice, 23 groups are generally used, the first 3 groups are used for subject familiarization tests, and the last 20 groups are used for formal tests. Therefore, after obtaining the optimized single digit based on the homogeneity optimization test results, it is necessary to construct a triplet library consisting of 23 groups of 3 digits according to the following steps:

[0071] Step (3.1) randomly combines the corrected audiometric materials, i.e., the ten numbers 0, 2-10, in groups of three, to obtain a total of 120 groups of initial triplets. However, this does not mean that all 120 groups are used for formal testing, because a qualified triplet library needs to meet three conditions: (1) frequency coverage, a single triplet can cover half or more of the speech sound frequency range (500-4000Hz); (2) balance of consonants, vowels, and tones, which is reflected in two aspects: ① the number of tested initials, finals, and tones is as large as possible; ② the tested initials, finals, and tones have as high a dynamic usage frequency as possible in Mandarin; (3) the number of occurrences of each number is as equal as possible. The next step is to reduce the 120 groups of triplets to the ideal triplet library according to the above rules:

[0072] Step (3.2), in accordance with frequency coverage:

[0073] Step (3.2.1), calculate the frequency range: the pronunciation of a number includes three parts: initial consonant, final consonant and tone (for example, 8, pronounced as eight, the initial consonant is b, the final consonant is a, and the tone is first tone), among which the initial consonant and final consonant are related to the frequency. Since the methods of calculating the frequency of initial consonants and final consonants are different, it is necessary to perform segmentation processing before calculating the frequency, that is, calculate the frequency of the initial consonant and final consonant parts of each number in a single initial triplet separately and perform weighted integration to form the frequency range of each number. Specifically, first import the speech material pre-processed in step (2) into the professional speech analysis software praat, and then mark the initial consonant and final consonant segments of each number pronunciation according to acoustic and perceptual clues. After marking the segments, read the values ​​of the first (Formant1, F1) formant peak and the second formant peak (Formant2, F2) of the final consonant segment to calculate the spectral center of gravity (Center of Gravity, COG) of the initial consonant segment. The spectral center of gravity refers to the concentrated area of ​​the strongest frequency in all critical bands. The spectral center of gravity is the average value of the entire frequency domain f and is calculated by |S(f)| p Weighted. The formula is as follows:

[0074]

[0075] After calculating the frequency, integrate the frequencies of the initial and final parts, such as Figure 3 and Figure 4 As shown, the frequency range of each number is formed.

[0076] Step (3.2.2), delete the group with incomplete frequency coverage (2, 5, 8), and 119 groups remain: Although the human ear hearing range is 0-20000Hz, digital triplet tone, as a speech audiometry method, is most concerned with the speech sound frequency range (500-4000Hz), so a set of triplet tone test listening materials should preferably cover the frequency range of 500-4000Hz. If less than half of the frequency range can be tested, it is considered to be poor material. For better quantification, it is now stipulated that the frequency score of a single triplet material that fully covers the speech sound frequency range (500-4000Hz) is recorded as 3 points, for example (3, 7, 9); covering half of the range (about 2250Hz) is recorded as 2 points, for example (0, 1, 6); covering obviously less than half (far less than 2250Hz) is recorded as 1 point, for example (2, 5, 8), and needs to be eliminated. That is, according to the preset speech sound frequency range (500-4000Hz), the initial triplets whose sum of the average value of the frequency range of each digit in each initial triplet are greater than or equal to the sum average value (2250Hz) of the upper limit (4000Hz) and lower limit (500Hz) of the preset speech sound frequency range are selected, and the scores are performed according to the preset scoring rules to form a triplet library that meets the frequency coverage requirements.

[0077] Step (3.3) meets the balance of phonetics, rhymes and tones:

[0078] Step (3.3.1), statistics of Mandarin tones: In addition to the frequency coverage evaluation dimension, the number of tones covered and the dynamic frequency of use are also important indicators for evaluating speech audiometry materials. A statistical table of digital tones is established based on the initials, finals and tones of each digital test. The statistics of the tones of the 10 numbers 0, 2-10 are shown in Table 1:

[0079] Table 1: Statistics of digital vowels and tones

[0080]

[0081] Note: Symbol In phonetics, it is used to record the zero initial consonant; although the finals of 4 and 10 are both written as pinyin i, they are actually pronounced differently. The i in 4 is a front vowel on the tip of the tongue, and the actual pronunciation is represented by the International Phonetic Alphabet as The i in 10 is a back vowel, and its actual pronunciation is represented by the International Phonetic Alphabet as

[0082] Step (3.3.2), 119 groups are deleted and then 4 groups that only test 1 initial consonant and 2 groups that only test 1 tone are deleted, and 113 groups are left: first, the quantity coverage, a group of triplet has at most 3 different initial consonants and at least 1 initial consonant - (3, 4, 5), (2, 3, 4), (0, 2, 6), (0, 5, 6); at most 3 different finals and at least 2 finals; at most 3 different tones and at least 1 tone - (3, 7, 8), (2, 4, 6), that is, the triplets that test the same initial consonant, the same final and the same tone are removed from the triplet library that meets the frequency coverage, and the triplet library to be optimized is obtained.

[0083] Step (3.3.3), dynamic usage frequency: refers to the distribution of speech in a word list or word entry with frequency parameters, which can better reflect the status of speech in actual language use than static usage frequency. Based on the statistical data of dynamic usage frequency of Mandarin consonants, finals and tones, the 7 initials, 9 finals and 4 tones involved in the pronunciation of numbers are sorted from high to low, and then assigned scores from high to low in turn, because the higher the dynamic usage frequency, the more commonly used it is in language use, the higher the score. For example, i is the most used among the 9 finals involved, so the score assigned is also the highest, reaching 9 points. According to the order of the usage frequency of initials, finals and tones from high to low, the scores are assigned in descending order, and the dynamic usage frequency statistics of Mandarin consonants, finals and tones and the corresponding score table are established, as shown in Table 2.

[0084] Table 2: Statistics on the dynamic usage frequency of Mandarin vowels, consonants and tones and their corresponding scores

[0085]

[0086] Step (3.4), score calculation: Based on step (3.1) and step (3.2), there are still 113 groups that meet the basic conditions. This step requires selecting the best 23 groups from these 113 groups to participate in the test. The optimal algorithm needs to use the scores assigned according to the frequency and tone conditions, that is, according to the scores of the frequency coverage triple sound library and the dynamic frequency statistics of Mandarin tone and the corresponding score table, generate an attribute table with numbers, pinyin, phonetic symbols, frequency, initial consonant usage frequency, final usage frequency and tone usage frequency, which is now organized as shown in Table 3:

[0087] Table 3: Attribute table

[0088]

[0089] The logic of the optimal algorithm is to normalize the seven key dimensions and calculate the sum of the scores, that is, to determine the sum of the scores of each triplet in the triplet library to be optimized according to the attribute table: let f be the frequency, s n is the number of initial consonants, s f is the frequency of initial consonants, yn is the number of finals, y f is the frequency of use of finals, t n is the number of tones, t f is the frequency of tone usage. Among them, frequency f and initial consonant usage frequency s f , frequency of use of finals f Frequency of harmony tone usage f Directly corresponds to the value of the above attribute table, the number of initials s n 、Number of finals s f The number of harmonic tones n Further calculation is required, corresponding to s f ,y f and t f The number of non-repeating elements.

[0090] For example, the attribute value corresponding to a set of triplets (0, 3, 4) is:

[0091] f=f 0 +f 3 +f 4 =2+3+3=8;s n = unique_count(s 0 ,s 3 ,s 4 )=unique_count

[0092] (4, 1, 1) = 2;

[0093] s f =s 0 +s 3 +s 4 =4+1+1=6;y n = unique_count(y 0 ,y 3 ,y 4 )=unique_count

[0094] (6, 5, 2) = 3;

[0095] y f =y 0 +y 3 +y 4 =6+5+2=13; t n = unique_count(t 0 , t 3 , t 4 )=unique_count

[0096] (3, 2, 4) = 3;

[0097] tf =t 0 +t 3 +t 4 =3+2+4=9.

[0098] The seven-dimensional attribute values ​​are obtained from the above. Since the score ranges of the seven dimensions are inconsistent, normalization processing is required. The maximum value normalization method is used in this invention. The maximum values ​​of the 113 groups of triplets are respectively recorded as f max ,s n_max ,s f_max ,y n_max ,y f_max , t n_max and t f_max , get the normalized dimension value:

[0099] (1) (2) (3) (4)

[0100] (5) (6) (7)

[0101] Finally, the total score of material quality is:

[0102] S=f norm +s n_norm +s f_norm +y n_norm +y f_norm +t n_norm +t f_norm

[0103] Among them, S, as the sum of the normalized scores, reflects the overall quality of the material. The higher the S value, the better the performance of the material in seven dimensions, namely, frequency, number of initials and usage frequency, number of finals and usage frequency, and number of tones and usage frequency. Therefore, the overall quality is higher. That is, the triplet library to be optimized is sorted according to the total score, and a preset number of triplets are selected to form the optimized triplet library.

[0104] Step (3.5), output the final 23 groups of triplets as audiometric materials: After calculating the scores, select the top 60 groups of scores. These 60 groups have met the requirements of frequency coverage and phonetic tone balance. Finally, they also need to meet the requirements of balanced digital frequency to avoid learning effects. Find 23 groups of digital triplets that meet the above three requirements to form a triplet library. The specific operation steps are as follows:

[0105] 1. Initialization:

[0106] Let T = {t1 ,t 2 ,…,t 60} is a collection of 60 triplets.

[0107] Let S = {s 1 ,s 2 ,…,s 60} is the score set for each group of triplets.

[0108] Let C be the set of selected triplets, initialized to the empty set.

[0109] Let F(d) be the frequency of number d in the selected triplet, initialized to 0.

[0110] 2. Selection process:

[0111] For each triplet t i ∈T, do the following:

[0112] Calculate triplet t i The frequency contribution of each number d in is ΔF(d).

[0113] Inspection i Whether the frequency constraint is satisfied: for all d∈t i , ensuring 6≤F(d)+ΔF(d)≤8

[0114] If t i If the frequency constraint is met, it is added to C and the frequency F(d)=F(d)+ΔF(d) is updated.

[0115] If |C|=23, stop selecting.

[0116] 3. Output results:

[0117] The number of selected triplet sets C is 23, and the frequency of occurrence of each number is between 6 and 8, which is the required triplet library. In this process, first initialize the relevant sets and variables. Then, traverse all triplets, evaluate whether they meet the frequency constraints, and select triplets with higher scores to join the final triplet library. Finally, we output 23 groups of triplets that meet the conditions, which constitute the final test materials, as shown in Table 4.

[0118] Table 4: Final digital triplet test library

[0119]

[0120] like Figure 5 As shown, the embodiment of the present invention also provides a method for using a digital triplet test library based on Chinese characteristics, using the final digital triplet test library constructed by the above method to perform a DTT test, and the specific steps are as follows:

[0121] like Figure 6 As shown, firstly, a Chinese digital triplet tone test system is constructed, including a laptop computer, a mouse, a professional headset and a test platform with a triplet tone library. Then, in the test platform, the triplet tone library is called for testing. The test platform can have multiple forms such as an online web version, an offline web version, an online application version, an offline application version, etc., as long as it has the final digital triplet tone test library created as described in the above method.

[0122] like Figure 7 As shown, considering the limited number of 23 groups and the existence of learning effect, in the formal test, the 23 groups did not appear in order but randomly. At the same time, the order of the three numbers in a single set could change, such as (2,8,10) could become (8,10,2), resulting in 23 groups of triplets.

[0123] During the DTT test, the first set of triplets is tested. The algorithm will determine that the answer is correct only when the patient gives a correct answer to all three numbers in the triplet.

[0124] If the answer is correct, the stimulus intensity is -2dB, and the second set of triplet tones is tested. If the answer is incorrect, the stimulus intensity is +2dB, and the second set of triplet tones is tested, until the 23rd set of triplet tones is tested. Based on the fluctuation of the intensity of multiple sets of triplet tones, the average signal-to-noise ratio will be calculated, that is, the digital recognition threshold under noise, in dBSNR, as the final hearing evaluation indicator.

Claims

1. A design method for a digital triplet test library based on Chinese characteristics, characterized in that: The following steps are involved: Obtain standardized digital audiometric materials, have multiple subjects conduct digital recognition experiments, and obtain the confusion matrix P based on the experimental results; According to the similarity valuation formula, the similarity valuation p′ is calculated for the data in the confusion matrix P. ij , combined with the perceptual distance formula, calculate the perceptual distance d between numbers ij Determine the number groups that are more likely to be confused, and retain some of the numbers in the number groups based on the number of initials and finals in the pronunciation. Add corresponding numbers based on the number, continuity and tone of the eliminated numbers to determine the audiometric materials; Conducting a homogeneity optimization test on the audiometric material, wherein the average speech recognition rate curve of each number in the audiometric material under different speech intensities or different signal-to-noise ratios is subjected to conformal interpolation, the speech recognition threshold of each number in the audiometric material before equalization and the average speech recognition threshold of all numbers in the audiometric material are determined, and the correction factor Factor is determined according to the correction factor formula, and the audiometric material is corrected according to the correction factor to obtain the corrected audiometric material; The corrected audiometric materials are randomly combined into multiple initial triplets in groups of three, the frequencies of the initial consonant and final parts of each number in a single initial triplet are calculated respectively and weighted integration is performed to form a frequency range of each number, and according to a preset speech sound frequency range, initial triplets whose sum average of the frequency range of each number in each initial triplet is greater than or equal to the sum average of the upper and lower bounds of the preset speech sound frequency range are selected, and they are scored according to a preset scoring rule to form a triplet library that meets the frequency coverage; A digital initial consonant, final consonant and tone statistical table is established according to each digital test, and triplets of the same initial consonant, the same final consonant and the same tone are removed from the triplet sound library that meets the frequency coverage, so as to obtain a triplet sound library to be optimized; According to the frequency of use of initials, finals and tones, scores are assigned in descending order from high frequency to low frequency, and a table of dynamic frequency statistics of Mandarin initials, finals and tones and corresponding scores is generated; according to the scores of the triple sound library that meets the frequency coverage and the table of dynamic frequency statistics of Mandarin initials, finals and tones and corresponding scores, an attribute table with numbers, pinyin, phonetic symbols, frequency, frequency of use of initials, frequency of use of finals and frequency of use of tones is generated; Determine the total score of each triplet in the triplet library to be optimized according to the attribute table, sort the triplet library to be optimized according to the total score, and select a preset number of triplets to form an optimized triplet library; The optimized triplet tone library is traversed, and the final digital triplet tone test library is screened out according to the preset balanced digital frequency and the total score of the triplet tones.

2. A design method for a digital triplet test library based on Chinese characteristics as claimed in claim 1, characterized in that: The similarity estimation formula is as follows: Among them, S ij It represents the probability of hearing the sound i as the sound j.

3. A design method for a digital triplet test library based on Chinese characteristics as claimed in claim 1, characterized in that: The perception distance formula is as follows: d ij =-ln(S ij )。 4. A design method for a digital triplet test library based on Chinese characteristics as claimed in claim 1, characterized in that: The correction factor formula is as follows: Among them, SRT i represents the speech recognition threshold before equalization of each digit in the audiometric material, indicating The average speech recognition threshold for all digits in the audiometric material.

5. The design method of a digital triplet test library based on Chinese characteristics as claimed in claim 1, characterized in that: Determining the total score of each triplet in the triplet library to be optimized according to the attribute table includes the following steps: Determine the frequency, initial consonant usage frequency, final consonant usage frequency and tone usage frequency of each number according to the attribute table, and calculate the number of initial consonants s of each triplet in the triplet library to be optimized n 、Number of finals s f The number of harmonic tones n , get the seven-dimensional attribute value of each triplet; The seven-dimensional attribute values ​​are normalized to their maximum values ​​and the sum is calculated to obtain the total score.

6. A design method for a digital triplet test library based on Chinese characteristics as claimed in claim 1, characterized in that: The preset speech sound frequency range is 500-4000 Hz.

7. A design method for a digital triplet test library based on Chinese characteristics as claimed in claim 1, characterized in that: The preset number is 23 groups.

8. A design method for a digital triplet test library based on Chinese characteristics as claimed in claim 1, characterized in that: The preset balanced digital frequency is greater than or equal to 6 and less than or equal to 8.

9. A method for using a digital triplet test library based on Chinese characteristics, characterized in that: The final digital triplet test library established by the design method of the digital triplet test library based on Chinese characteristics as claimed in claim 1 comprises the following steps: The order of triplets in the final digital triplet test library is disrupted, and the order of the numbers in the triplets is disrupted to obtain multiple groups of triplets in a disrupted order; During the test, the patient was first tested based on the first set of triplets. The judgment standard was that when the patient gave a correct answer to all three numbers in the triplets, the answer was considered correct. If the answer is correct, the intensity of the stimulus sound is reduced by a preset value, and the second set of triplet sounds is continued for testing; if the answer is incorrect, the intensity of the stimulus sound is increased by a preset value, and the second set of triplet sounds is continued for testing, until all sets of triplet sounds are tested; The signal-to-noise ratio is calculated according to the fluctuation of the test intensity to obtain the final hearing evaluation index.

Images