JP2534242B2 - Pattern comparison method - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a pattern comparison method, and more particularly, to a pattern comparison method for performing pattern matching in pattern recognition.

2. Description of the Related Art The most common pattern recognition method is a method using pattern matching. This method seeks the similarity between a standard pattern registered in advance and an input unknown input, and determines which of the registered patterns the unknown input is closest to. When this method is actually applied, there is a problem that ambient noise is mixed in the target pattern. For example, when considering speech recognition, if a sudden noise is made in the surroundings during recognition, or if the opening / closing sound of the lips during utterance is included in the speech, noise is generated in the pattern matching unit. Since it is compared with the standard pattern as it is, correct similarity is not required. Also, if it is hypersensitive to noise, the above-mentioned noise addition is likely to occur, but if sensitivity to noise is lowered, noise will not be extracted, but on the other hand, it causes a situation where a part of voice is missing. More often. For example, in the case of a word such as stop that is likely to be generated by a consonant alone, the last / p / may not be extracted and may be omitted.

FIG. 9 is a diagram for explaining an example of collating a stop as described above, in which (a) is a standard pattern,
It is assumed that (b) is the input pattern. The standard pattern (a) has exactly the pattern of the whole word,
In the pattern (b), / p / is missing, and at the end of the pattern, / p / of the pattern (a) is matched with / o / of the pattern (b). For this reason, the similarity between the two becomes small, which causes erroneous recognition. This kind of mishandling can be prevented by using end-point-free dynamic programming, but the pattern at the beginning and the end may be missing, or noise may be added, which is a method with a large amount of calculation. However, there is a drawback that the amount of calculation is further increased.

In view of the above situation, the present applicant has previously proposed a pattern comparison method that can perform correct pattern comparison while using a pattern with a small amount of calculation and a pattern with a noise added thereto. .

FIG. 10 is a block diagram for explaining an example of the pattern comparison method previously proposed by the present applicant, in which 1 is a voice input unit, 2 is a power measurement unit, 3 is a feature amount conversion unit, 4 is a silent position measuring unit, 5 is a silent position (end) determination unit, 6 is a mark addition unit, 7 is a dictionary, 8 is a mark determination unit, 9 is a pattern deletion unit, 10 is a pattern matching unit, and the voice is first The power of the voice is measured in the process of being input and converted into a feature amount. The voice input unit 1 can be executed by a microphone, and the power measurement unit 2
Can be realized by detecting the output of the bandpass filter only in the voice band. Various conversion methods of the characteristic amount are known, but any method such as a power spectrum that can be easily extracted by a bandpass filter group may be used. From the measured power level, measure where the silent section (the minimum point of voice energy) is located in the voice. For example, a mark is added to a pattern (feature pattern) whose feature amount is converted depending on whether or not this silent position is within 100 ms from the end of the voice. On the other hand, it is checked whether or not the pattern in the dictionary has the same mark as the characteristic pattern of the extraction destination, and if the marks are the same, the pattern matching section is normally entered. A method for creating a dictionary pattern is not particularly described here, but a similar method is used to mark in advance depending on whether or not there is silence at the end of the dictionary pattern. If the mark in the dictionary is different from the mark in the input, the silence from the one with silence to the end of the pattern is deleted, and matching is performed using the pattern from the beginning to the silence. For example, as shown in FIG. 11, consider the case where the pattern (a) in the dictionary is normal and the end of the pattern (b) that is the input pattern is missing, and in this case, the pattern (a) is at the end. It has a mark that there is no sound, and pattern (b) becomes a mark that there is no sound. Therefore, a pattern (c) in which / p / at the end of the pattern (a) which is a dictionary pattern is deleted is created, and the patterns (b) and (c) are associated with each other and collated. It should be noted that the above description has been made regarding the loss of the end of the voice, but the present invention is not limited to the end, and the same thing can be done at the beginning.

FIG. 12 is a diagram showing an example of pattern comparison when there is a dropout at the beginning of the voice. In the figure, 11 is a silent position (beginning) determination part, and other operations similar to those of the example shown in FIG. The same reference numerals as those in FIG. 10 are attached to the portions marked with. Thus, in this case, depending on whether or not there is silence within the beginning of 100 ms, the beginning portion is canceled as in the above example, and then verification is performed. In this case, the pattern of the dictionary is considered to be normal and complete, and the input speech has the noise added by the opening and closing of the lips or the noise generated suddenly.

FIG. 13 is a diagram for explaining an example of such a case. For example, when uttering "move" / ido /, if noise n is attached at the beginning like pattern (a), this noise Since a silent section is generated after, the silent position measuring unit 4 and the silent position (starting) determining unit 11 confirm that this silent section is generated within 100 ms at the beginning, and add a mark with a silent section. On the other hand, since there is no silence in the dictionary pattern (b), the portion closer to the beginning than the silent section of the pattern (a) is deleted by the pattern deleting unit 9, and after being deleted, the pattern matching unit 10 matches. The means for matching is not limited, but it is desirable that the amount of calculation that linearly expands or contracts the pattern length is small. In this case, needless to say, the time length of the pattern (a) is changed after the initial silence. Further, the silent section measuring unit 4 records environmental noise in a state where no voice is input, and when the level of the voice when the voice is input can be regarded as noise, the silent section signal is set as the silent section signal. The position measuring unit 11 may measure the position by comparing with the signal of the signal line indicating the rising and falling of the voice section.

Thus, the pattern comparison method proposed by the applicant of the present application seeks a threshold value such as 100 ms in order to find the energy minimum in the vicinity of the beginning. Is marked with no minimal part, and if one pattern has a minimal value in 90 ms, it is removed and the pattern is collated.

FIG. 14 is a diagram showing an example in that case. In this case, it is determined that the pattern (a) has no minimum power near the end, and the pattern (b) is determined to have a minimum power. ) / P / is removed and the pattern (a) is matched.

OBJECT The present invention has been made in view of the above-mentioned circumstances, and in particular, it is an object of the present invention to provide a pattern comparison method that enables accurate pattern matching even if the power minimum is slightly deviated.

Structure In order to achieve the above-mentioned object, when comparing two patterns, it is checked whether or not there is a minimum of energy in the vicinity of the start end or the end of the pattern, and from the minimum of the pattern having the minimum to the start or In the pattern comparison method that compares both patterns after removing up to the end,
(1) When the numbers of the minimum portions in both patterns do not match, deleting from the first or last minimum value of the pattern having a large number of minimum portions to the pattern end, and collating, or
(2) It is checked whether or not there is a minimum energy in the input signal. If there is a minimum energy, the ratio of the minimum position to the pattern length is obtained, and the ratio is close to the standard pattern ratio and If there is an energy minimum only near the end points of the pattern, check that both patterns have or do not have a minimum, or (3) check whether there is an energy minimum in the input signal. , If it exists, the ratio of the minimum position to the pattern length is calculated, and it is a value close to the ratio of the standard pattern, and the pattern length from the minimum to the pattern end is less than a certain value and one pattern end point When there is an energy minimum only in the vicinity, matching is performed assuming that there is a minimum in either pattern, or (4) the energy of the input signal is reduced. If the unknown input pattern and the standard pattern have different numbers, the pattern length from the minimum to the silence closest to the end point of the larger one is subtracted from the pattern length of its own. It is characterized in that the other pattern lengths are compared, and when the value is closer than before the subtraction, the minimum to the end point are deleted and then the comparison is performed. Examples of the present invention will be described below.

FIG. 1 is a flow chart for explaining an embodiment of the present invention, and FIG. 2 is a block diagram showing an example of an electric circuit used for implementing the embodiment shown in FIG. 20
Is a microphone, 21 is a voice segment cutout unit, 22 is a bandpass filter (BPF), 23 is a register, 24 is a voice power measuring device, 25 is a comparison circuit, 26 is a threshold value generation unit, 27 is a time counter, and 28 is a comparison circuit. , 29 is a silent section counter, 30
Is a mark comparison unit, 31 is a changeover switch, 32 is a dictionary, 33
Is a pattern deletion unit, 34 is a collation unit, and 35 is a result display unit. In this embodiment, when comparing two patterns, if there is a minimum energy in the vicinity of the beginning or end of the pattern, a pattern having a minimum In the pattern comparison method that compares and collates both patterns after removing from the minimum to the beginning or the end, when the number of minimums in both patterns does not match, the first or last minimum of the pattern with many minimums From the pattern end to the pattern end, the minimum energy value of the voice is described as a silent portion in the voice. First, after the voice is input, the power is measured, it is checked whether there is a silent section, and the number thereof is obtained. Next, the number of silent intervals in the standard pattern in the dictionary is compared with the number, and if the numbers match, the collation is performed as usual. The matching method is not particularly limited, and any desired method may be used.
Only when the numbers do not match, the preparation for pattern deletion is started. That is, the power of the voice input from the microphone 20 is measured by the voice power measuring unit 24 to measure the silent position, and a different mark is attached depending on whether the silent position is within the first 100 ms, for example. The dictionary pattern registered with marks in this way is taken out, and it is judged whether or not it is the same mark as the inputted pattern. If they are the same mark, the changeover switch 31 is set to the side a and the pattern matching is started as usual. On the other hand, when the marks are different, the changeover switch 31 is set to the b side as shown in the figure, and the pattern deleting unit 33 deletes from the first or last local minimum value of the pattern having a large number of local minimum parts to the pattern end, and then, The collation unit 34 collates. Therefore, an erroneous pattern due to a silent position shift is eliminated, and accurate matching can be performed.

FIG. 3 is a flow chart for explaining another embodiment of the present invention, and FIG. 4 is a block diagram of an electric circuit used for implementing the embodiment shown in FIG. A comparison unit, 37 is a silence number generation unit, 38 is a changeover switch, 39 is a calculation unit, 40 is a comparison unit, 41 is a multiplication constant setting unit, and 42 is a changeover switch. Others are the same as those in the embodiment shown in FIG. The parts that operate are given the same reference numerals as in FIG. Thus, in this embodiment, when comparing two patterns, it is checked whether or not there is a minimum energy in the vicinity of the start end or the end of the pattern, and if there is a minimum in only one pattern, In a pattern comparison method that compares and collates both patterns after removing the minimum to the beginning or the end of the pattern having the minimum, it is checked whether the minimum energy exists in the input signal, and if there is, the pattern If the ratio of the minimum position to the length is found and it is a value that is close to the ratio of the standard pattern and there is an energy minimum only near the end points of one pattern,
The comparison is performed by assuming that there is a minimum or no minimum in either pattern, but here, the minimum value of energy will be described as a silent part in the voice. First,
After the voice is input, the power is measured to check whether there is a silent section, and if there is, find its position. If the position of the silent section is within 100 ms at the beginning, mark that there may be noise at the beginning of the voice,
At the same time, the number of silent sections is calculated. When there are two or more silent sections in either direction, the difference in the position ratio between the silent sections
If it is within 10%, switch 42 from the position shown in the figure to perform normal verification without passing through the pattern deletion section.
42 is set to the position shown in the drawing, and the pattern having the most silent sections is guided to the deletion section and then collated. In other words, the two patterns whose positions in the silent section are 110 ms and 90 ms are considered to have no noise added and no pattern is missing, and normal matching is performed. That is, it is determined that the other pattern is missing and the other pattern is collated. Here, the silent position ratio is the pattern length fms and the silent position f ′.
It is defined as ms by f '/ f. When the pattern length is long and the number of silent sections is large, the error becomes large only by the ratio. If there is a local minimum of energy in the input signal, if it exists, the ratio of the local minimum position to the pattern length is calculated, and it is a value close to the ratio of the standard pattern. If the pattern lengths up to are less than a fixed value and the energy minimum is present only near one of the pattern end points, it is assumed that there is a minimum in either pattern or not, and matching is performed.

FIG. 5 is a flow chart for explaining an embodiment in the case of performing the pattern matching by obtaining the ratio of the minimum position to the pattern length when the minimum energy exists as described above, and FIG. FIG. 3 is a block diagram showing an example of an electric circuit used for implementing the embodiment shown in the figure. In this embodiment, the absolute position or time until silence is checked after checking the difference in silent position ratio, and the difference is calculated. The pattern is deleted only when is longer than 200 ms. Of course, 10% and 200 ms described here are examples, and the present invention is not limited to this, and a value that can be regarded as a neighborhood may be selected. This does not necessarily require the ratio of the silent section, and can be performed by the method shown in FIG. 7, for example.

FIG. 7 is a flow chart for explaining another embodiment of the present invention, and FIG. 8 is a block diagram showing an example of an electric circuit used for implementing the embodiment shown in FIG. ,
Parts having the same functions as those of the embodiment shown in FIG. 6 are designated by the same reference numerals as those in FIG. Thus, in this embodiment, when comparing two patterns, it is checked whether or not there is a minimum energy in the vicinity of the start or end of the pattern, and the minimum to the start or end of the pattern having the minimum is taken. In the pattern comparison method that compares the patterns of both after removing them, it is checked whether or not there is a minimum energy in the input signal, and when the number is different between the unknown input pattern and the standard pattern, The pattern length from the minimum point near the end point to the end point is subtracted from the pattern length of its own and the other pattern length is compared, and when the value is closer than before subtraction, the minimum point to the end point is deleted and then the comparison is performed. It is a thing. That is, when there is silence at the beginning of the input pattern and the mark is different from the dictionary pattern to be matched, first, the length Fb obtained by subtracting the pattern length from the beginning of the input pattern to the silence from the overall pattern length Fa is calculated. Whether the pattern length F ₁ is close to Fa or Fb is checked, and only when the pattern length F ₁ is close to Fb, the pattern length F ₁ is included in the deletion portion.

Effect As is clear from the above description, according to the present invention, even if it is difficult to distinguish between voice and noise, it is automatically determined, and the necessity of noise removal is checked before the removal is determined. The accuracy of matching is improved.

[Brief description of drawings]

FIG. 1 is a flow chart for explaining an embodiment of the present invention, FIG. 2 is a block diagram of an electric circuit used for implementing the embodiment shown in FIG. 1, and FIG. 4 is a flow chart for explaining another embodiment of the present invention, FIG. 4 is a block diagram of an electric circuit used for implementing the embodiment shown in FIG. 3, and FIG. 5 is another embodiment of the present invention. 6 is a flow chart for explaining, FIG. 6 is a block diagram of an electric circuit used for implementing the embodiment shown in FIG. 5, and FIG. 7 is a flow chart for explaining another embodiment of the present invention. FIG. 8 is a block diagram of an electric circuit used for implementing the embodiment shown in FIG. 7, FIG. 9 is a pattern diagram for explaining the technical background of the present invention, and FIG. 10 is the present application. FIG. 11 is a block diagram for explaining an example of a pattern comparison method previously proposed by a person. 10 is a pattern matching diagram for explaining the operation of FIG. 10, FIG. 12 is a block diagram for explaining another example, and FIG. 13 is a pattern matching diagram for explaining the operation of the example shown in FIG. , FIG. 14 is a diagram showing an example of a voice pattern to which the present invention is applied. 20 …… Microphone, 21 …… Voice section cutout unit, 22…
… Bandpass filter (BPF), 23 …… Register, 24…
… Voice power measurement unit, 25 …… Comparison circuit, 26 …… Threshold value generation unit, 27 …… Time counter, 28 …… Comparison circuit, 29 …… Silence section counter, 30 …… Mark comparison unit, 31 …… Selection switch , 32 …… Dictionary, 33 …… Pattern deletion part, 34 …… Verification part, 35 …… Result display part, 36 …… Comparison circuit, 37 …… Silence number generation part, 38 …… Changeover switch, 39 …… Calculation Section, 40 …… Comparison circuit, 41 …… Multiplication constant setting section, 42 …… Changeover switch, 43 …… Comparison circuit, 44 …… Comparison circuit, 45 …… Predetermined time (2
00ms) Generator, 46, 47 ...... Calculator, 49 ... Changeover switch.

Claims (4)

(57) [Claims] 1. When comparing two patterns, it is checked whether or not there is a minimum energy in the vicinity of the start or end of the pattern, and after removing from the minimum to the start or end of the pattern having the minimum, In the pattern comparison method that compares both patterns, when the number of local minimums in both patterns does not match, the pattern from the first or last local minimum value of the pattern with a large number of local minimums to the pattern end may be deleted and matched. Characteristic pattern comparison method. 2. When comparing two patterns, it is checked whether or not there is a minimum energy in the vicinity of the start or end of the pattern, and after removing from the minimum of the pattern having the minimum to the start or end, In the pattern comparison method that compares both patterns, check whether the input signal has a minimum energy, and if there is, calculate the ratio of the minimum position to the pattern length and compare it with the standard pattern ratio. The pattern comparison method is characterized in that when there is a minimum energy in the vicinity of the end points of one of the patterns, it is assumed that there is a minimum in either pattern, and matching is performed. 3. When comparing two patterns, it is checked whether or not there is a minimum energy in the vicinity of the start or end of the pattern, and after removing from the minimum of the pattern having the minimum to the start or end, In the pattern comparison method that compares both patterns, check whether the input signal has a minimum energy, and if there is, calculate the ratio of the minimum position to the pattern length and compare it with the standard pattern ratio. If the pattern length from the minimum of both to the pattern end is less than a certain value and the energy minimum is only near one of the pattern end points, it is assumed that both patterns have a minimum or not. A pattern comparison method characterized by matching. 4. When comparing two patterns, it is checked whether or not there is a minimum energy in the vicinity of the beginning or end of the pattern, and after removing from the minimum of the pattern having the minimum to the beginning or end, In the pattern comparison method that compares both patterns, it is checked whether or not there is a minimum energy in the input signal, and when the number is different between the unknown input pattern and the standard pattern, it is closer to the end point of the larger one. A pattern characterized by comparing the value obtained by subtracting the pattern length from the minimum to the silence from its own pattern length with the other pattern length, and when the value is closer than before subtraction, the minimum to end points are deleted and then the pattern is compared. Comparison method.