JP2000322080A - Speech recognition processor and speech recognition processing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a speech recognition processor and a speech recognition processing method capable of recognizing speech with high speed and a high recognition rate. SOLUTION: This speech recognition processor 100 is equipped with matching processors 3 and 4, an initial matching buffer 7#1, a matching buffer 6, and a beam search processor 8. The matching processor 4 performs a matching process while judging whether a frame of an initial segment will be an initial edge or not. The result is stored in the initial matching buffer 7#1. In the matching processor 3, a matching process of a frame positioned in other than the initial segment is performed. The result is stored in the matching buffer 6. The beam search processor 8 performs a beam search process on the basis of the information in the matching buffer 6.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a speech recognition processing apparatus and a speech recognition processing method, and more particularly to a speech recognition processing apparatus and a speech recognition processing method for performing a matching process with a standard pattern.

[0002]

FIG. 8 shows a conventional speech recognition processor 800.
FIG. 2 is a block diagram showing an outline of the configuration of FIG. The arrow shown in FIG. 8 indicates the flow of processing for one frame. Referring to FIG. 8, voice recognition processing device 800 includes microphone 40, A / D
It includes a converter 41, an acoustic analyzer 42, a voice section detector 43, a matching processor 44, a matching buffer 45, a beam search processor 46, and a standard pattern storage 47.

In the speech recognition processing device 800, the microphone 40
Sampled audio input through
The digital conversion is performed in the D converter 41. The A / D converter 41 divides the digital data into units for performing voice recognition processing (frame division) and transmits the data to the acoustic analyzer 42 in frame units. In the acoustic analyzer 42, the A / D converter 4
The data structure of the frame received from 1 is converted into a format in which the matching processor 44 performs a matching process. The voice section detector 43 detects a voice section from the output of the acoustic analyzer 42. The data determined as a voice section by the voice section detector 43 is transmitted to the matching processor 44. The matching processor 44 performs a matching process between the data of the voice section and the standard pattern stored in the standard pattern storage 47 in advance.

[0006] The result of the matching process is stored in a matching buffer 45. The matching buffer 45 stores, for example, (frame number of input pattern, frame number of matched standard pattern, similarity) as accumulated information.

FIG. 9 shows a speech recognition processing device 80 shown in FIG.
FIG. 9 is a conceptual diagram for explaining the operation of the “0”. In FIG. 9, the vertical axis represents the frame of the standard pattern, and the horizontal axis represents the frame of the input voice (referred to as an input pattern) in time series. The number of standard patterns is N (standard patterns 1 to N). Note that, in the figure, a symbol 25 represents a matching path.

In the speech recognition processor 800, the start frame on the input pattern side (the beginning of the matching process) is fixed, and the start frame on the standard pattern side is moved (frames 1 to Fh). Hereinafter, a section including a plurality of frames that can be the start frame is referred to as an end point free section.

Referring to FIG. 8, beam search processor 46
Performs a beam search process using the result of the matching buffer 45. In the beam search process, it is determined based on the similarity stored in the matching buffer 45 whether or not the standard pattern used in the matching process is to be recognized. For example, a standard pattern whose similarity stored in the matching buffer 45 is higher than a predetermined threshold (dissimilarity is high) is not recognized. The calculation speed is increased by interrupting the matching process.

FIG. 10 shows a conventional speech recognition processing apparatus 900.
FIG. 2 is a block diagram showing an outline of the configuration of FIG. The difference between the speech recognition processing device 900 shown in FIG. 10 and the speech recognition processing device 800 shown in FIG. 8 is that the speech recognition processing device 900 includes a matching processor 54 instead of the matching processor 44 and does not include the beam search processor 46. is there.

[0009] Speech recognition processing apparatus 900 determines a speech section in the same procedure as speech recognition processing apparatus 800.
The matching processor 54 performs a matching process with a standard pattern for a frame determined to be a voice section.
At this time, as shown in FIG. 11, the frames 1 to Fi are set as the end point free sections on the input pattern side, and the frames 1 to Fh are set as the end point free sections on the standard pattern side.

[0010]

As described above, in the conventional speech recognition processing apparatus 800, the start frame on the input pattern side to be subjected to the matching process is fixed, and the standard pattern side is included in the end point free section. Are set as candidates for the start frame. This is because moving the start frame on both the input pattern side and the standard pattern side complicates the processing.

However, if noise is input during voice input, the voice section detector 43 may erroneously detect a voice section. In such a case, since the start frame on the input pattern side is fixed, the result of the erroneous detection affects the matching process in the initial stage, and there is a problem that the recognition rate is reduced.

In the conventional speech recognition processing apparatus 900, an end point free section is provided on both the standard pattern side (frames 1 to Fh) and the input pattern side (frames 1 to Fi), and recognition is performed by moving the start frame. The rate is improving.

However, it is difficult to install the beam search processor 45 for performing the sequential processing.
For this reason, there has been a problem that the processing speed becomes slow due to the huge amount of calculation.

The present invention has been made to solve such a problem, and an object of the present invention is to provide a speech recognition processing device and a speech recognition processing method having a high recognition rate while speeding up speech recognition processing. To provide.

[0015]

According to a first aspect of the present invention, there is provided a voice recognition processing apparatus for performing voice recognition by determining whether an input voice pattern is similar to a standard pattern stored in advance. A recognition processing device, a storage device for storing a standard pattern, a voice input device for dividing an input voice pattern into frames, and outputting the frame in frame units, and determining whether a frame output from the voice input device belongs to a voice section. A detector that detects whether the frame belongs to the initial section of the voice section, and a section that determines whether the frame output by the detector belongs to the initial section. A first process of performing matching processing with a frame of a corresponding standard pattern while determining whether or not the frame corresponds to a start frame of a matching path.
And a second matching processor that performs a matching process on a frame of a standard pattern corresponding to a frame determined to belong to a section other than the initial section by the section determiner.

According to a second aspect of the present invention, there is provided the voice recognition processing apparatus according to the first aspect, wherein the first matching processor determines a result of the matching processing of a frame immediately before the input frame. A first processor for performing a matching process with a frame of a corresponding standard pattern, a second processor for performing a matching process between an input frame and a start frame of the corresponding standard pattern, and a result of the first processor. A comparator for comparing the result with the second processor to determine a matching path.

According to a third aspect of the present invention, there is provided the voice recognition processing apparatus according to the second aspect, wherein a corresponding standard pattern is subjected to voice recognition based on a result of the matching processing in the second matching processor. And a second matching processor that stops the matching process with the corresponding standard pattern based on the determination result of the inspector.

A speech recognition processing device according to a fourth aspect is the speech recognition processing device according to the second aspect, wherein a corresponding standard pattern is converted into a speech based on a result of the matching process in the initial section after the matching process in the initial section. An inspection device for determining whether or not to perform recognition;
Of the matching processor stops the matching process with the corresponding standard pattern based on the determination result of the inspection device.

According to a fifth aspect of the present invention, there is provided a voice recognition processing method for performing voice recognition by determining whether an input voice pattern is similar to a standard pattern stored in advance. A voice input step of dividing an input voice pattern into frames and outputting the frame in frame units; a detection step of detecting whether a frame output by the voice input step belongs to a voice section; and determining the voice section in the detection step. A matching process step of performing a matching process on the selected frame, wherein the matching process step includes a section determining step of determining whether a frame output by the detecting step belongs to an initial section of the voice section, and a section determining step. Frame is determined to belong to the initial section A first matching processing step of performing frame and matching processing of corresponding standard pattern while it is determined whether or not corresponding to the start frame of the road,
A second matching processing step of performing a matching process on a frame determined to belong to a section other than the initial section in the section determination step with a frame of a corresponding standard pattern.

According to a sixth aspect of the present invention, there is provided the voice recognition processing method according to the fifth aspect, wherein the first matching processing step is performed in accordance with a result of the matching processing of the frame immediately before the input frame. A first processing step of performing matching processing with a frame of a corresponding standard pattern, a second processing step of performing matching processing between an input frame and a start frame of the corresponding standard pattern, and a result of the first processing step. Comparing the result of the second processing step to determine a matching path.

According to a seventh aspect of the present invention, there is provided the voice recognition processing method according to the sixth aspect, wherein a corresponding standard pattern is subjected to voice recognition based on a result of the matching processing in the second matching processing step. In the second matching processing step, the matching processing with the corresponding standard pattern is stopped based on the determination result in the inspection step.

The voice recognition processing method according to claim 8 is the voice recognition processing method according to claim 6, wherein a corresponding standard pattern is subjected to voice recognition based on a result of the matching processing in the second matching processing step. A first inspection step of determining whether or not the first frame has been processed. The first matching processing step includes a step of determining whether or not the matching processing of the last frame in the initial section has been completed, and a step of determining whether the last frame in the initial section has been completed. A second inspection step of determining whether or not the corresponding standard pattern is to be subjected to speech recognition based on a result of the matching process in the initial section after the end of the matching process;
In the matching process step, the matching process with the corresponding standard pattern is stopped based on the determination result of the first inspection step or the determination result of the second inspection step.

[0023]

[First Embodiment] FIG. 1 is a block diagram showing an outline of a configuration of a speech recognition processing apparatus 100 according to a first embodiment of the present invention. Voice recognition processing device 80
The same functions as 0 and 900 are denoted by the same reference numerals and symbols, and description thereof will be omitted.

Referring to FIG. 1, speech recognition processing device 100
Is a microphone 40, an A / D converter 41, an acoustic analyzer 42,
Section position determiner 2, matching processors 3 to 4, standard pattern storage 5, matching buffer 6, initial matching buffers 7 # 1 to 7 # 2, and beam search processor 8
Is provided.

As described above, the input voice pattern is divided into frames via the microphone 40 and the A / D converter 41, and the data structure is converted in the acoustic analyzer 42. Then, the voice section detector 43 detects a voice section. The section position determiner 2 determines whether the frame output from the voice section detector 43 is a frame of an initial section. The frames in the initial section are processed by the matching processor 4.
Other frames are processed by the matching processor 3 respectively.

FIG. 2 is a diagram for explaining the initial section. In FIG. 2, symbol 9 represents a voice.
The initial section means a certain section (time t0 to t4) including the beginning of the voice, as shown in FIG. In the case of FIG. 2, the frame F0 (time t0 to t
2), F1 (time t1 to t3), F2 (time t2 to t)
4) and a frame F3 (time t3 to t5).

The frames F0 to F3 are processed by the matching processor 4, and other frames, for example, frames F4 (time t4 to t6) and F5 (time t5 to t5)
7), F6 (time t6 to t8), F7 (time t7 to...)
Are processed by the matching processor 3.

Referring to FIG. 1, the result of matching in matching processor 4 is an initial matching buffer 7 #.
1 is stored. The result of the matching process in the matching processor 3 is stored in the matching buffer 6. The standard patterns 1 to N to be matched are stored in the standard pattern storage 5.

FIG. 3 is a diagram showing a matching path of an input pattern with respect to a standard pattern. The horizontal axis represents frames in the input pattern, and the vertical axis represents frames in the standard pattern in time series. Symbol 25 indicates a matching path, and indicates a positional relationship between an input pattern and a standard pattern that provides an optimum matching result.

On the input pattern side, frames 1 to Fi are
On the standard pattern side, frames 1 to Fh are each defined as an end point free section. At this time, it is assumed that the initial section includes K frames (K> i). The method of setting the initial section is determined based on a balance between the recognition rate and the processing speed.

Next, the operation of the speech recognition processing device 100 according to the first embodiment of the present invention will be described with reference to FIG. FIG. 4 is a flowchart for explaining the operation of the speech recognition processing device 100. Voice input is performed by the microphone 40 and the A / D converter 41 (step S1). Voice analyzer 4
2, a voice analysis process is performed (step S2). A voice section is detected by the voice section detector 43 (step S3). When the voice section detector 43 determines that the voice section is a voice section, the section position determination unit 2 determines whether or not the voice section is an initial section (initial stage) (step S).
4).

The frames included in the initial section are transmitted to the matching processor 4. The matching processor 4 has an end point discriminator (has a function of judging whether or not it corresponds to the first end)
(Step S5). Frames included in sections other than the initial section are transmitted to the matching processor 3. The matching processor 3 performs a normal matching process (step S6). That is, a limited path is searched for from a frame that matches the frame of the previous input pattern, and a frame that matches the frame of the current input pattern is found.

The result of the matching processing with end point determination (step S5) is stored in the initial matching buffer 7 # 1. The result of the normal matching process (step S6) is stored in the matching buffer 6. The beam search processor 8 performs a beam search process using the data stored in the matching buffer 6 (Step S7). The above steps S1 to S7 are repeated, and when a speech section is no longer detected, speech recognition is determined using the results of the matching buffer 6 and the initial matching buffer 7 # 1 (step S8). This recognition determination result is output on a display (not shown) (step S9).

FIG. 5 is a flow chart for explaining the operation in the matching processing with end point discrimination (matching processor 4) shown in FIG. Referring to FIG. 5, first, a normal matching process is performed (step S10). For example, the matching process is performed between the frame of the standard pattern and the frame located in the vicinity of the previously matched standard pattern and the frame of the input pattern. This result is stored in the initial matching buffer 7 # 1 (step S11).

Next, the target of the matching process is set to frame 1 (start frame) in the standard frame (step S12), and the matching process is performed (step S1).
3). The result of the matching process is stored in the initial matching buffer 7 # 2.

Subsequently, the results of the initial matching buffers 7 # 1 and 7 # 2 are compared (step S15). The result stored in the initial matching buffer 7 # 1 is
If the result is better than the result stored in the initial matching buffer 7 # 2 (that is, the matching path in step S10 is better than that in step S13), the matching processing for the frame is ended.

On the other hand, the result stored in the initial matching buffer 7 # 2 is better than the result stored in the initial matching buffer 7 # 1 (ie, step S10).
If the matching path in step S13 is better), the information in the initial matching buffer 7 # 2 is copied to the initial matching buffer 7 # 1.

An example will be described. For example, last time
Input pattern frame 5 and standard pattern frame 4
And are matched. Matching path
[1, 2, 2, 3, 4] and the similarity is 30. In step S10, a matching process is performed between the input pattern frame 6 and the standard pattern frame 4 and neighboring frames (frames 5, 6). Thus, for example, it is assumed that the frame 6 of the standard pattern matches more than the other frames (similarity becomes 36).

Next, in step S13, a matching process is performed between frame 1 of the standard pattern and frame 6 of the input pattern. In step S14, step S13
Is compared with the result (similarity) of step S10. For example, when the result of step S13 is the similarity 10, it is assumed that the standard pattern frame 6 is the best matching. If the result of step S13 is similarity 2, the frame 1 in the standard pattern is determined to be the best matching, and the information in the initial matching buffer 7 # 1 is replaced with the initial matching buffer 7 #.
Rewrite using 2.

As described above, in the initial section, the matching process is performed while determining which frame is the start frame (initial end). The result is stored in the initial matching buffer 7 # 1.

Referring to FIG. 4, normal matching processing is performed on frames included in sections other than the initial section. This result is stored in the matching buffer 6.
The beam search processor 8 performs a beam search process based on the similarity stored in the matching buffer 6. At this time, since the result of the matching process in the initial section requires a huge amount of calculation, it is excluded from the beam search process in the beam search processor 8.

The beam search processor 8 recognizes a standard pattern having a clearly high degree of dissimilarity as a recognition target. The certification result of the beam search processor 8 is transferred to the matching buffer. The matching buffer 6
A flag indicating whether or not to perform the matching process is provided, and the certification result of the beam search processor 8 is written in this flag.
The matching processor 3 performs a matching process while checking the flag. For those that are not recognized,
Stop the matching process. Note that the matching processor 3 may be directly controlled based on the result of the beam search processor 8.

The operations in steps S1 to S7 are performed until the operation goes out of the voice section. Outside the voice section, the beam search processor 8 takes into account the result of the matching process in the initial section stored in the initial matching buffer 7 # 1 and the result of the matching process stored in the matching buffer 6.
In step (2), a standard pattern representing the result of the best matching process is selected and output from the standard patterns to be recognized.

As described above, according to the speech recognition processing apparatus 100 in the first embodiment of the present invention, in the initial section, the matching process is performed while determining whether or not the input pattern side corresponds to the first end. (An end point free section is provided on the input pattern side). This makes it possible to increase the recognition rate irrespective of the accuracy of voice section detection.

Further, by performing beam search processing outside the initial section, it is possible to narrow down the target, so that high-speed processing can be performed on a plurality of standard patterns, and a high recognition rate can be obtained. It becomes possible.

[Second Embodiment] FIG. 6 is a block diagram showing an outline of a configuration of a speech recognition processing device 200 according to a second embodiment of the present invention. Speech recognition processing device 1 shown in FIG.
The same functions as those of 00 are denoted by the same reference numerals and symbols, and description thereof is omitted. The difference between the speech recognition processing device 200 and the speech recognition processing device 100 is that the beam search controller 2
6 inclusive.

The beam search controller 26 determines that the result of the matching process for the last frame of the initial section has been stored in the initial matching buffer 7 # 1, and performs the matching process stored in the initial matching buffer 7 # 1. The beam search processor 8 is controlled to perform a beam search process based on a value representing the result. The value representing the result of the matching process is, for example, an average value, a center value, or the like of the similarities in each frame of the stored input pattern.

Next, the operation of the speech recognition processing device 200 according to Embodiment 1 of the present invention will be described with reference to FIG. FIG. 7 is a flowchart for explaining the operation of the speech recognition processing device 200. The same steps as those of the operation of the speech recognition processing device 100 shown in FIG. 4 are denoted by the same reference numerals, and description thereof will be omitted.

As described above, the section position determiner 2 determines whether or not the input voice belongs to the initial section (initial stage) (step S4). As described above, the matching process with the end point determination is performed for the initial section (step S5). And the beam search controller 2
6, it is determined that the result of the matching process for the last frame of the initial section is stored in the initial matching buffer 7 # 1 (step S20). When the processing for the last frame has been completed (step S2
1) The beam search processor 8 performs a beam search process based on a value representing the result of the matching process stored in the initial matching buffer 7 # 1.

When the value representing the result of the matching process in the initial section is equal to or larger than the predetermined threshold, the beam search processor 8 excludes the standard pattern from recognition. The result of the beam search processing is stored in the matching buffer 6 as described above.

A normal matching process is performed on the frame of the input pattern in a section other than the initial section (step S6). At this time, the standard pattern that is not recognized as a result of the matching processing in the initial section or the result of the matching processing in the section other than the initial section is
Not subject to matching processing. Then, the beam search processor 8 performs a beam search process in response to the result of the matching process in a section other than the initial section (step S7).

As described above, according to the speech recognition processing apparatus 200 in the second embodiment of the present invention, it is possible to improve the recognition rate by making the result of the matching in the initial section the target of the beam search processing. . In addition, since the recognition is not performed for the one that is already bad in the initial section, the recognition processing can be performed at high speed.

The embodiments disclosed this time are to be considered in all respects as illustrative and not restrictive. The scope of the present invention is defined by the terms of the claims, rather than the description above, and is intended to include any modifications within the scope and meaning equivalent to the terms of the claims.

[0054]

As described above, according to the speech recognition processing apparatus according to the first and second aspects, it is determined whether or not the section (initial section) including the beginning of the input speech can be an end point (start frame). The matching process is performed while judging whether or not. For this reason, it is possible to increase the recognition rate irrespective of the accuracy of the detection of the voice section in the initial stage.

A speech recognition processing device according to a third aspect is the speech recognition processing device according to the second aspect, and further includes, in a section other than the initial section, a corresponding standard pattern to be recognized based on a result of the matching processing. (Beam search processing) is performed. This makes it possible to narrow down the targets, so that high-speed processing can be performed on a plurality of standard patterns, and a high recognition rate can be obtained.

The speech recognition processing device according to a fourth aspect is the speech recognition processing device according to the second aspect, and in particular, determines whether a corresponding standard pattern is to be recognized based on a matching process result in an initial section. Inspection (beam search processing) is performed. This makes it possible to perform a high-speed recognition process because a poorly matched object in the initial section is excluded from recognition targets.

Further, according to the speech recognition processing method according to the fifth and sixth aspects, it is determined whether or not a section (initial section) including the beginning of the input speech can be an end point (start frame). While performing the matching process. For this reason, it is possible to increase the recognition rate irrespective of the accuracy of the detection of the voice section in the initial stage.

According to a seventh aspect of the present invention, there is provided the voice recognition processing method according to the sixth aspect, wherein a corresponding standard pattern is recognized based on a result of the matching process in a section other than the initial section. (Beam search processing) is performed. This makes it possible to narrow down the target, so that high-speed processing can be performed on a plurality of standard patterns, and a high recognition rate can be obtained.

The speech recognition processing method according to the eighth aspect is the speech recognition processing method according to the sixth aspect. In particular, a method for determining whether a corresponding standard pattern is to be recognized based on a matching processing result in an initial section. Inspection (beam search processing) is performed. This makes it possible to perform a high-speed recognition process because a poorly matched object in the initial section is excluded from recognition targets.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating an outline of a configuration of a speech recognition processing device 100 according to a first embodiment of the present invention.

FIG. 2 is a diagram illustrating an initial section.

FIG. 3 is a diagram illustrating a matching path of an input pattern with respect to a standard pattern.

FIG. 4 is a flowchart for explaining the operation of the speech recognition processing device 100.

FIG. 5 is a flowchart for explaining the operation in the matching processing with end point determination (matching processor 4) shown in FIG. 4;

FIG. 6 is a block diagram illustrating an outline of a configuration of a speech recognition processing device 200 according to a second embodiment of the present invention.

FIG. 7 is a flowchart for explaining the operation of the speech recognition processing device 200.

FIG. 8 is a block diagram showing a configuration of a conventional speech recognition processing device 800.

FIG. 9 is a conceptual diagram for explaining a matching path of an input pattern with respect to a standard pattern in the speech recognition processing device 800.

FIG. 10 is a block diagram showing a configuration of a conventional speech recognition processing device 900.

FIG. 11 is a conceptual diagram for explaining a matching path of an input pattern with respect to a standard pattern in the speech recognition processing device 900.

[Explanation of symbols]

2 Section position determiner, 3, 4 matching processor, 5
Standard pattern storage, 6 matching buffers, 7♯1
♯7♯2 Initial matching buffer, 8,26 beam search processor, 26 beam search controller, 40 microphone, 41 A / D converter, 42 acoustic analyzer, 43 voice section detector, 100-200 voice recognition processor .

Claims (8)

[Claims] 1. A voice recognition processing device for performing voice recognition by determining whether an input voice pattern is similar to a standard pattern stored in advance, and a storage unit storing the standard pattern A voice input device that divides the input voice pattern into frames and outputs the frame in frame units; a detector that detects whether a frame output from the voice input device belongs to a voice section; and A section determiner that determines whether the output frame belongs to an initial section of a voice section; and for the frame determined to belong to the initial section in the section determiner, the frame is determined as a start frame of a matching path. A first matching processor that performs matching processing with a frame of a corresponding standard pattern while determining whether or not the frame is applicable; A speech recognition processing device, comprising: a second matching processor that performs a matching process on a frame of a standard pattern corresponding to a frame determined to belong to a section other than an initial section by a section determiner. 2. The first matching processor according to claim 1, further comprising: a first processor that performs a matching process on a frame of a corresponding standard pattern according to a result of the matching process on a frame immediately before the input frame. A second processor for performing a matching process between the frame and a start frame of the corresponding standard pattern; and comparing a result of the first processor with a result of the second processor to determine a matching path. The speech recognition processing device according to claim 1, further comprising a comparator. 3. An inspection device for determining whether or not a corresponding standard pattern is to be subjected to the speech recognition based on a result of the matching process in the second matching process device, wherein the second matching process is performed. 3. The speech recognition processing device according to claim 2, wherein the device stops the matching process with the corresponding standard pattern based on the determination result of the inspection device. 4. A tester for determining whether or not a corresponding standard pattern is to be subjected to the speech recognition based on a result of the matching process in the initial section after the matching process in the initial section is completed, The speech recognition processing device according to claim 2, wherein the second matching processor stops the matching process with the corresponding standard pattern based on the determination result of the inspection device. 5. A voice recognition processing method for performing voice recognition by determining whether or not an input voice pattern is similar to a standard pattern stored in advance, wherein the input voice pattern is divided into frames. A voice input step of outputting in frame units; a detection step of detecting whether or not the frame output by the voice input step belongs to a voice section; and a frame determined to be a voice section in the detection step. A matching process step of performing a matching process, wherein the matching process step includes: a section determining step of determining whether the frame output by the detecting step belongs to an initial section of a voice section; For the frame determined to belong to the initial section, the frame is A first matching processing step of performing matching processing with a frame of a corresponding standard pattern while determining whether or not the frame corresponds to a start frame of the matching path; and the section determining step determines that the frame belongs to a section other than the initial section. A second matching process step of performing a matching process on the frame with a corresponding standard pattern frame. 6. The first matching processing step includes: a first processing step of performing matching processing with a frame of a corresponding standard pattern according to a result of matching processing of a frame immediately before the input frame; A second processing step of performing a matching process between the frame and a start frame of a corresponding standard pattern; and comparing a result of the first processing step with a result of the second processing step to determine a matching path. 6. The speech recognition processing method according to claim 5, comprising a comparing step. 7. An inspection step for determining whether or not a corresponding standard pattern is to be subjected to the speech recognition based on a result of the matching processing in the second matching processing step, wherein the second matching processing is performed. 7. The speech recognition processing method according to claim 6, wherein in the step, the matching process with the corresponding standard pattern is stopped based on a determination result in the inspection step. 8. A first method for judging whether or not a corresponding standard pattern is to be subjected to the speech recognition based on a result of the matching processing in the second matching processing step.
Further comprising an inspection step, wherein the first matching processing step is a step of determining whether or not the matching processing of the last frame in the initial section has been completed; and A second inspection step of determining whether or not a corresponding standard pattern is to be subjected to the voice recognition based on a result of the matching processing in the initial section; and wherein the second matching processing step includes the first inspection step 7. The speech recognition processing method according to claim 6, wherein the matching process with the corresponding standard pattern is stopped based on the determination result of the second inspection step.