English consonant p identification method based on interpolation wavelet and tower-type decomposition

Abstract

The invention discloses an English consonant p identification method based on interpolation wavelet and tower decomposition, and belongs to the field of signal analysis. A six-order spline interpolation wavelet # imgabs0 # and a corresponding scale function psi (omega) are used, corresponding interpolation low-pass and high-pass filters are built, then a low-pass filter coefficient {alpha n} and a high-pass filter coefficient {beta n} are used, a dual interpolation filter coefficient {phi n} of the low-pass filter is calculated, and the dual interpolation filter coefficient {phi n} of the low-pass filter is calculated. Performing convolution operation on the core component {Xn} of the audio signal and the dual filter coefficient {phi n}, and calculating low-frequency wavelet coefficients {cj} in a plurality of scale spaces; discrete Fourier transform is carried out on the low-frequency wavelet coefficient {gamma m}, and feature points of input audio signal energy and the distribution range of the feature points are obtained. And the feature information is finely compared with the feature information of the standard consonant 'p', so that the purpose of identifying and classifying the components of the audio signal is achieved. Compared with the traditional method based on Fourier transform and classic wavelet transform, the method provided by the invention has the advantages that the calculation speed, the accuracy and the stability of the algorithm are improved.

Description

Technical Field

[0001] The present invention relates to the field of signal analysis, and in particular to a method for extracting and identifying signal features of an English phoneme consonant "p" based on interpolation wavelet and pyramidal decomposition. Background Art

[0002] As one of the key technologies in the field of artificial intelligence, the development of speech recognition technology is of great significance to improving the naturalness and efficiency of human-computer interaction. Globally, the speech processing algorithm market has long been monopolized by a few foreign technology giants, which have controlled the market dominance with their patent barriers and technological innovations. This monopoly not only limits the development space of other countries and enterprises in this field, but also increases dependence on external technologies. With its deep professional knowledge and innovative spirit, Mr. Zhang's team invented a patent for speech processing algorithm designed from the bottom up, which provides a new choice for domestic speech processing technology. As an advanced signal processing method, interpolation wavelet transform has shown its unique advantages in speech recognition. Interpolation wavelet transform can extract the time-frequency characteristics of speech signals and provide richer information for recognition algorithms; multi-scale analysis of speech signals is helpful for accurate segmentation of phonemes; by performing threshold processing in the wavelet domain, background noise can be effectively suppressed and the accuracy of speech recognition can be improved. At the same time, it overcomes the difficulties that traditional Fourier transform cannot take into account time-frequency information and classical wavelet transform has large computational complexity and low accuracy. Summary of the invention

[0003] In order to reduce the amount of computational work in audio signal processing and improve detection accuracy, Professor Zhang Zhiguo innovatively proposed a method for identifying the English phoneme consonant "p" based on interpolation wavelet and pyramidal decomposition. On the basis of traditional wavelet transform, interpolation technology was introduced to improve the accuracy and resolution of the transform, thereby achieving the effect of optimizing the calculation.

[0004] The technical solution of the present invention is: a method for recognizing English consonant p based on interpolation wavelet and pyramidal decomposition, the method comprising: the specific steps are as follows:

[0005] Step 1: Sample the input audio signal represented as a set {x n};

[0006] Step 2: Take the set {x n The element with the maximum absolute value of} is denoted as x m ;

[0007] x m =|x n | max (1)

[0008] Step 3: Read the audio element set {x n} is normalized to its maximum absolute value and recorded as {x * n}

[0009] x * n =x n / x m (2)

[0010] Step 4: Select a threshold THR = 0.003 and perform the normalized audio element set {x * n}Threshold processing is performed,

[0011] If |x * n |≤THR then x * n =0, and record the resulting array as {X n};

[0012] Step 5: The corresponding interpolation low-pass filter is constructed by the sixth-order spline interpolation wavelet scaling function ψ(ω), and its coefficient is recorded as {α n}, And n∈[-15,15];

[0013] Step 6: By {α n}Calculate the corresponding interpolation high-pass filter coefficient {β n}, And n∈[-15,15];

[0014] Step 7: By {α n} and {β n}, And n∈[-15,15], the corresponding low-pass filter's dual filter interpolation filter coefficient is recorded as {φ n}, And n∈[-15,15];

[0015] Step 8: Calculate the low-frequency wavelet coefficients in the zeroth space based on the following formula Where k is an array {X n} data length, φ k-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, The data length is m;

[0016]

[0017] Step 9: Based on the calculation results of the previous step, calculate the wavelet low-frequency decomposition coefficients in the first space φ j-2nis the dual filter interpolation filter coefficient corresponding to the low-pass filter, and after calculation, m is updated to The data length;

[0018]

[0019] Step 10: Based on the calculation results of the previous step, calculate the wavelet low-frequency decomposition coefficients in the second space φ j-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, and after calculation, m is updated to The data length;

[0020]

[0021] Step 11: Based on the calculation results of the previous step, calculate the wavelet low-frequency decomposition coefficients in the third space φ j-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, and after calculation, m is updated to The data length;

[0022]

[0023] Step 12: Based on the calculation results of the previous step, calculate the wavelet low-frequency decomposition coefficients in the fourth space φ j-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, and after calculation, m is updated to The data length;

[0024]

[0025] Step 13: Based on the calculation results of the previous step, calculate the wavelet low-frequency decomposition coefficients in the fifth space φ j-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, and after calculation, m is updated to The data length;

[0026]

[0027] Step 14: Based on the calculation results of the previous step, calculate the wavelet low-frequency decomposition coefficients in the sixth space φ j-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, and after calculation, m is updated to The data length;

[0028]

[0029] Step 15: Based on the calculation results of the previous step, calculate the wavelet low-frequency decomposition coefficients in the seventh space φ j-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, and after calculation, m is updated to The data length;

[0030]

[0031] Step 16: Keep the wavelet coefficients of the low-frequency signals in the fifth, sixth, and seventh spaces

[0032] Step 17: Perform discrete Fourier transform on the obtained wavelet decomposition coefficients;

[0033]

[0034] Step 18: For {τ 5}{τ 6}{τ 7} are processed according to formula (14) and recorded as arrays {τ 5*}{τ 6*}{τ 7*}, called feature points:

[0035]

[0036] Among them, ξ t (n) is an array {ξ t}, ξ t (n-1) is the array {ξ t}, the n-1th number in t (n+1) is the array {ξ t}, where n is the length of the array;

[0037] Step 19: Use the following formula to calculate the sum of the feature points of the fifth, sixth and seventh spaces respectively;

[0038]

[0039] Step 20: Determine the sum S of the fifth space eigenvalues 5 Is it less than 0.01? If so, proceed to the next step; otherwise, determine that the input is not a "p" phoneme;

[0040] Step 21: Note {τ 5 *} The non-zero element interval is set A1, and the distribution range of the feature points in the fifth space is set B1. If A1 is a subset of B1, that is, A1∈B1, then further judgment is required. If A1 is not a subset of B1, that is, Then it is judged that the input is not a "p" phoneme;

[0041] Step 22: Determine S6 Is it less than 0.05? If so, further judgment is required. Otherwise, it is judged that the input is not a "p" phoneme.

[0042] Step 23: Note {τ 6 *} The non-zero element interval is set A2, and the feature point distribution range of the sixth space is set B2. If A2 is a subset of B2, that is, A2∈B2, then further judgment is required. If A2 is not a subset of B2, that is, Then it is judged that the input is not a "p" phoneme;

[0043] Step 24: Determine S 7 If it is less than 0.05, further judgment is required, otherwise the input is judged not to be a "p" phoneme;

[0044] Step 25: Note {τ 7 *} The non-zero interval is set A3, and the distribution range of the feature points in the seventh space is set B3. If A3 is a subset of B3, that is, A3∈B3, then further judgment is required. If A3 is not a subset of B3, that is, Then it is judged that the input is not a "p" phoneme;

[0045] Step 26: Output the judgment result "the audio is a "p" phoneme" or "the audio is not a "p" phoneme" according to steps 20 to 26.

[0046] Furthermore, the coefficients of the interpolation low-pass filter in step 5 are {α n}, its values ​​are shown in the following table:

[0047] <![CDATA[α -15 ]]> <![CDATA[α -14 ]]> <![CDATA[α -13 ]]> <![CDATA[α -12 ]]> <![CDATA[α -11 ]]> <![CDATA[α -10 ]]> <![CDATA[α -9 ]]> <![CDATA[α -8 ]]> -0.0005 0.0000 0.0012 0.0000 0.0027 0.0000 0.0063 0.0000 <![CDATA[α -7 ]]> <![CDATA[α -6 ]]> <![CDATA[α -5 ]]> <![CDATA[α -4 ]]> <![CDATA[α -3 ]]> <![CDATA[α -2 ]]> <![CDATA[α -1 ]]> <![CDATA[α0 <!-- 3 -->]]> 0.0147 0.0000 0.0343 0.0000 0.0840 0.0000 0.3099 0.5000 <![CDATA[α1]]> <![CDATA[α2]]> <![CDATA[α3]]> <![CDATA[α4]]> <![CDATA[α5]]> <![CDATA[α6]]> <![CDATA[α7]]> <![CDATA[α8]]> 0.3099 0.0000 -0.0840 0.0000 0.0343 0.0000 -0.0147 0.0000 <![CDATA[α9]]> <![CDATA[α 10 ]]> <![CDATA[α 11 ]]> <![CDATA[α 12 ]]> <![CDATA[α 13 ]]> <![CDATA[α 14 ]]> <![CDATA[α 15 ]]> 0.0063 0.0000 -0.0027 0.0000 0.0012 0.0000 -0.0005 .

[0048] Furthermore, the interpolation high-pass filter coefficients {β n}, its values ​​are shown in the following table:

[0049]

[0050]

[0051] Furthermore, the dual filter interpolation filter coefficients in step 7 are denoted as {φ n}, whose values ​​are shown in the following table:

[0052] <![CDATA[φ -15 ]]> <![CDATA[φ -14 ]]> <![CDATA[φ -13 ]]> <![CDATA[φ -12 ]]> <![CDATA[φ -11 ]]> <![CDATA[φ -10 ]]> <![CDATA[φ -9 ]]> <![CDATA[φ -8 ]]> -0.0060 -0.0073 0.0091 0.0111 -0.0138 0.0166 0.0212 0.0246 <![CDATA[φ -7 ]]> <![CDATA[φ -6 ]]> <![CDATA[φ -5 ]]> <![CDATA[φ -4 ]]> <![CDATA[φ -3 ]]> <![CDATA[φ -2 ]]> <![CDATA[φ -1 ]]> <![CDATA[φ0]]> -0.0331 -0.0356 0.0539 0.0485 -0.0999 -0.0598 0.3162 0.5644 <![CDATA[φ1]]> <![CDATA[φ2]]> <![CDATA[φ3]]> <![CDATA[φ4]]> <![CDATA[φ5]]> <![CDATA[φ6]]> <![CDATA[φ7]]> <![CDATA[φ8]]> 0.3162 -0.0598 0.0999 0.0485 0.0539 -0.0356 -0.0331 0.0246 <![CDATA[φ9]]> <![CDATA[φ 10 ]]> <![CDATA[φ 11 ]]> <![CDATA[φ 12 ]]> <![CDATA[φ 13 ]]> <![CDATA[φ 14 ]]> <![CDATA[φ 15 ]]> 0.0212 -0.0166 -0.0138 0.0111 0.0091 -0.0073 -0.0060 .

[0053] The present invention uses the sixth-order spline interpolation wavelet and the corresponding scaling function ψ(ω), build the corresponding interpolation low-pass and high-pass filters, and then use the low-pass filter coefficients {α n} and high-pass filter coefficients {βn}, calculate the dual interpolation filter coefficients of the low-pass filter {φ n}, using Mallat tower decomposition technology, the core components of the audio signal {X n} and the dual filter coefficients {φ n} performs convolution operation to calculate the low-frequency wavelet coefficients {c j Next, the low-frequency wavelet coefficients {γ m}Discrete Fourier transform is performed to obtain the characteristic points and distribution range of the input audio signal energy. By carefully comparing these characteristic information with the characteristic information of the standard consonant "p", the purpose of component recognition and classification of the audio signal is achieved. Compared with the traditional methods based on Fourier transform and classical wavelet transform, the phoneme recognition technology proposed by the present invention has improved calculation speed, accuracy and algorithm stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0055] The present invention proposes an English consonant p recognition method based on interpolation wavelet and pyramidal decomposition, the method comprising:

[0056] First, input the audio and use the sampling and holding technique to convert it into a discrete signal for analysis {x n}; Normalize the collected audio signal and obtain standardized audio data through threshold processing {X n}; A sixth-order spline wavelet was selected The corresponding sixth-order spline interpolation wavelet scaling function ψ(ω) is derived through mathematical operations; these wavelet scaling functions are used to calculate the coefficients {α n} and {β n}, and then derive the dual interpolation filter coefficients of the low-pass filter {φ n}; Using the Mallat pyramid decomposition method, the core components of the audio signal {X n} and the dual filter coefficients {φ n} performs convolution operation to calculate the low-frequency wavelet coefficients {γ m}; Perform discrete Fourier transform on the wavelet coefficients to obtain the spectrum of the input signal, and identify the feature points and their distribution characteristics as the signal recognition conditions; by comparing with the feature points and distribution characteristics of the phoneme consonant "p" audio, determine whether the input is the consonant "p";

[0057] The technical solution of the present invention is: a method for recognizing English consonant p based on interpolation wavelet and pyramidal decomposition, the method comprising: the specific steps are as follows:

[0058] Step 1: Sample the input audio signal represented as a set {x n};

[0059] Step 2: Take the set {x n The element with the maximum absolute value of} is denoted as x m ;

[0060] x m =|x n | max (1)

[0061] Step 3: Read the audio element set {x n} is normalized to its maximum absolute value and recorded as {x * n}

[0062] x * n =x n / x m (2)

[0063] Step 4: Select a threshold THR = 0.003 and perform the normalized audio element set {x * n}Threshold processing is performed,

[0064] If |x * n |≤THR, then x * n =0, and record the resulting array as {X n};

[0065] Step 5: The corresponding interpolation low-pass filter is constructed by the sixth-order spline interpolation wavelet scaling function ψ(ω), and its coefficient is recorded as {α n}, And n∈[-15,15], its value is shown in the following table:

[0066]

[0067]

[0068] Step 6: By {α n}Calculate the corresponding interpolation high-pass filter coefficient {β n}, And n∈[-15,15], its value is shown in the following table:

[0069] <![CDATA[β -15 ]]> <![CDATA[β -14 ]]> <![CDATA[β -13 ]]> <![CDATA[β -12 ]]> <![CDATA[β -11 ]]> <![CDATA[β -10 ]]> <![CDATA[β -9 ]]> <![CDATA[β -8 ]]> 0.0000 -0.0012 0.0000 0.0027 0.0000 -0.0064 0.0000 0.0148 <![CDATA[β -7 ]]> <![CDATA[β -6 ]]> <![CDATA[β -5 ]]> <![CDATA[β -4 ]]> <![CDATA[β -3 ]]> <![CDATA[β -2 ]]> <![CDATA[β -1 ]]> <![CDATA[β0]]> 0.0000 -0.0345 0.0000 0.0842 0.0000 -0.3100 0.5000 -0.3100 <![CDATA[β1]]> <![CDATA[β2]]> <![CDATA[β3]]> <![CDATA[β4]]> <![CDATA[β5]]> <![CDATA[β6]]> <![CDATA[β7]]> <![CDATA[β8]]> 0.0000 0.0842 0.0000 -0.0345 0.0000 0.0148 0.0000 -0.0064 <![CDATA[β9]]> <![CDATA[β 10 ]]> <![CDATA[β 11 ]]> <![CDATA[β 12 ]]> <![CDATA[β 13 ]]> <![CDATA[β 14 ]]> <![CDATA[β 15 ]]> 0.0000 0.0027 0.0000 -0.0012 0.0000 0.0005 0.0000

[0070] Step 7: By {α n} and {β n}, And n∈[-15,15], the corresponding low-pass filter's dual filter interpolation filter coefficient is recorded as {φ n}, And n∈[-15,15], its value is shown in the following table:

[0071]

[0072]

[0073] Step 8: Calculate the low-frequency wavelet coefficients in the zeroth space based on the following formula Where k is an array {X n} data length, φ k-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, The data length is m;

[0074]

[0075] Step 9: Based on the calculation results of the previous step, calculate the wavelet low-frequency decomposition coefficients in the first space φ j-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, and after calculation, m is updated to The data length;

[0076]

[0077] Step 10: Based on the calculation results of the previous step, calculate the wavelet low-frequency decomposition coefficients in the second space φ j-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, and after calculation, m is updated to The data length;

[0078]

[0079] Step 11: Based on the calculation results of the previous step, calculate the wavelet low-frequency decomposition coefficients in the third space φ j-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, and after calculation, m is updated to The data length;

[0080]

[0081] Step 12: Based on the calculation results of the previous step, calculate the wavelet low-frequency decomposition coefficients in the fourth space φ j-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, and after calculation, m is updated to The data length;

[0082]

[0083] Step 13: Based on the calculation results of the previous step, calculate the wavelet low-frequency decomposition coefficients in the fifth space φ j-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, and after calculation, m is updated to The data length;

[0084]

[0085] Step 14: Based on the calculation results of the previous step, calculate the wavelet low-frequency decomposition coefficients in the sixth space φ j-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, and after calculation, m is updated to The data length;

[0086]

[0087] Step 15: Based on the calculation results of the previous step, calculate the wavelet low-frequency decomposition coefficients in the seventh space φ j-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, and after calculation, m is updated to The data length;

[0088]

[0089] Step 16: Keep the wavelet coefficients of the low-frequency signals in the fifth, sixth, and seventh spaces

[0090] Step 17: Perform discrete Fourier transform on the obtained wavelet decomposition coefficients;

[0091]

[0092] Step 18: For {τ 5}{τ 6}{τ 7} are processed according to formula (14) and recorded as arrays {τ 5*}{τ 6*}{τ 7*}, called feature points:

[0094]

[0095] Among them, ξ t (n) is an array {ξ t}, ξ t (n-1) is the array {ξ t}, the n-1th number in t (n+1) is the array {ξ t}, where n is the length of the array;

[0096] Step 19: Calculate the sum of all feature points;

[0097]

[0098] Step 20: Determine the sum S of the fifth space eigenvalues 5 Is it less than 0.01? If so, proceed to the next step; otherwise, determine that the input is not a "p" phoneme;

[0099] Step 21: Note {τ 5* The non-zero element interval is set A1, and the distribution range of the feature points in the fifth space is set B1. If A1 is a subset of B1, that is, A1∈B1, then further judgment is required. If A1 is not a subset of B1, that is, Then it is judged that the input is not a "p" phoneme;

[0100] Step 22: Determine S 6 Is it less than 0.05? If so, further judgment is required. Otherwise, it is judged that the input is not a "p" phoneme.

[0101] Step 23: Note {τ 6* The non-zero element interval is set A2, and the feature point distribution range of the sixth space is set B2. If A2 is a subset of B2, that is, A2∈B2, then further judgment is required. If A2 is not a subset of B2, that is, Then it is judged that the input is not a "p" phoneme;

[0102] Step 24: Determine S 7 If it is less than 0.05, further judgment is required, otherwise the input is judged not to be a "p" phoneme;

[0103] Step 25: Note {τ 7* The non-zero interval is set A3, and the distribution range of the feature points in the seventh space is set B3. If A3 is a subset of B3, that is, A3∈B3, then further judgment is required. If A3 is not a subset of B3, that is, Then it is judged that the input is not a "p" phoneme;

[0104] Step 26: Output the judgment result "the audio is a "p" phoneme" or "the audio is not a "p" phoneme" according to steps 20 to 21.

Claims

1. A method for recognizing English consonant p based on interpolation wavelet and pyramidal decomposition, the method comprising: The specific steps are as follows: Step 1: Sample the input audio signal represented as a set {x n }; Step 2: Take the set {x n The element with the maximum absolute value of} is denoted as x m ; x m =x nmax (1) Step 3: Read the audio element set {x n } is normalized to its maximum absolute value and recorded as {x * n } x * n =x n / x m (2) Step 4: Select a threshold THR = 0.003 and perform the normalized audio element set {x * n }Threshold processing is performed, If |x * n |≤THR, then x * n =0, and record the resulting array as {X n }; Step 5: The corresponding interpolation low-pass filter is constructed by the sixth-order spline interpolation wavelet scaling function ψ(ω), and its coefficient is recorded as {α n }, And n∈[-15,15]; Step 6: By {α n }Calculate the corresponding interpolation high-pass filter coefficient {β n }, And n∈[-15,15]; Step 7: By {α n } and {β n }, And n∈[-15,15], the corresponding low-pass filter's dual filter interpolation filter coefficient is recorded as {φ n }, And n∈[-15,15]; Step 8: Calculate the low-frequency wavelet coefficients in the zeroth space based on the following formula Where k is the data length of the array {Xn}, φ k-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, The data length is m; Step 9: Based on the calculation results of the previous step, calculate the wavelet low-frequency decomposition coefficients in the first space φ j-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, and after calculation, m is updated to The data length; Step 10: Based on the calculation results of the previous step, calculate the wavelet low-frequency decomposition coefficients in the second space φ j-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, and after calculation, m is updated to The data length; Step 11: Based on the calculation results of the previous step, calculate the wavelet low-frequency decomposition coefficients in the third space φ j-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, and after calculation, m is updated to The data length; Step 12: Based on the calculation results of the previous step, calculate the wavelet low-frequency decomposition coefficients in the fourth space φ j-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, and after calculation, m is updated to The data length; Step 13: Based on the calculation results of the previous step, calculate the wavelet low-frequency decomposition coefficients in the fifth space φ j-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, and after calculation, m is updated to The data length; Step 14: Based on the calculation results of the previous step, calculate the wavelet low-frequency decomposition coefficients in the sixth space φ j-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, and after calculation, m is updated to The data length; Step 15: Based on the calculation results of the previous step, calculate the wavelet low-frequency decomposition coefficients in the seventh space φ j-2n is the dual filter interpolation filter coefficient corresponding to the low-pass filter, and after calculation, m is updated to The data length; Step 16: Keep the wavelet coefficients of the low-frequency signals in the fifth, sixth, and seventh spaces Step 17: Perform discrete Fourier transform on the obtained wavelet decomposition coefficients; Step 18: For {τ 5 }{τ 6 }{τ 7 } are processed according to formula (14) and recorded as arrays {τ 5* }{τ 6* }{τ 7* }, called feature points: Among them, ξ t (n) is an array {ξ t }, ξ t (n-1) is the array {ξ t }, the n-1th number in t (n+1) is the array {ξ t }, where n is the length of the array; Step 19: Use the following formula to calculate the sum of the feature points of the fifth, sixth and seventh spaces respectively; Step 20: Determine the sum S of the fifth space eigenvalues 5 Is it less than 0.01? If so, proceed to the next step; otherwise, determine that the input is not a "p" phoneme; Step 21: Note {τ 5 *} The non-zero element interval is set A1, and the distribution range of the feature points in the fifth space is set B1. If A1 is a subset of B1, that is, A1∈B1, then further judgment is required. If A1 is not a subset of B1, that is, Then it is judged that the input is not a "p" phoneme; Step 22: Determine S 6 Is it less than 0.05? If so, further judgment is required. Otherwise, it is judged that the input is not a "p" phoneme. Step 23: Note {τ 6 *} The non-zero element interval is set A2, and the distribution range of the feature points in the sixth space is set B2. If A2 is a subset of B2, that is, A2∈B2, then further judgment is required. If A2 is not a subset of B2, that is, Then it is judged that the input is not a "p" phoneme; Step 24: Determine S 7 If it is less than 0.05, further judgment is required, otherwise the input is judged not to be a "p" phoneme; Step 25: Note {τ 7 *} The non-zero interval is set A3, and the distribution range of the feature points in the seventh space is set B3. If A3 is a subset of B3, that is, A3∈B3, then further judgment is required. If A3 is not a subset of B3, that is, Then it is judged that the input is not a "p" phoneme; Step 26: Output the judgment result "the audio is a "p" phoneme" or "the audio is not a "p" phoneme" according to steps 20 to 26.

2. The English consonant p recognition method based on interpolation wavelet and pyramidal decomposition as claimed in claim 1, characterized in that: The coefficients of the interpolation low-pass filter in step 5 are {α n }, its values ​​are shown in the following table: 。 3. The English consonant p recognition method based on interpolation wavelet and pyramidal decomposition as claimed in claim 1, characterized in that: In step 6, the interpolation high-pass filter coefficients {β n }, its values ​​are shown in the following table: 。 4. The English consonant p recognition method based on interpolation wavelet and pyramidal decomposition as claimed in claim 1, characterized in that: The dual filter interpolation filter coefficients in step 7 are denoted as {φ n }, whose values ​​are shown in the following table: 。

Images