JP5347808B2 - Similar voice list generation apparatus, similar voice list generation method, and similar voice list generation program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a similar voice list creating device providing a part in which frequently appearing voice contents are recorded, beforehand to a user. <P>SOLUTION: A CPU 11 sequentially segments a voice block of a predetermined time length from a voice data as a voice block for comparison origin, and sequentially segments a voice block of a predetermined length for comparing it with the voice block for comparison origin, as a voice block for comparison destination. The CPU 11 calculates the feature amount of the voice block for comparison origin, and the feature amount of the voice block for comparison destination, and a distance between both the feature amounts is calculated. Distance information in which the calculated distance is associated with positional information of the voice block for comparison origin and the voice block for comparison destination, is stored in a RAM 13. When comparison of all voice blocks are finished, the CPU 11 obtains positional information of the voice block in which voice is similar, on the basis of the distance information stored in the RAM 13, and creates a similar voice block list in which the position information is grouped. <P>COPYRIGHT: (C)2011,JPO&amp;INPIT

Description

  The present invention relates to a similar sound list generation device, a similar sound, and a similar sound list generating device for extracting a data portion having similar sound from the sound data and generating a list in which the position information in the sound data is grouped by similar data portions. The present invention relates to a list generation method and a similar voice list generation program.

  For example, there are many cases where the content of a meeting or the like is recorded by a recording device such as an IC recorder, and the recorded content is heard later and used for work or the like.

  In such a case, conventionally, a person who has not participated in the meeting or the like has to listen to the recording content roughly equivalent to the recording time in order to grasp important points in the meeting or the like. Met.

  For example, Patent Literature 1 proposes a speech similarity calculation device for the purpose of automatically calculating the content similarity between voice messages or automatically classifying the content according to the content.

JP-A-10-105188

  However, the technique proposed in Patent Document 1 cannot solve the above-described conventional user inconvenience, and it is a fact that a new technique is desired to be proposed.

  The present invention has been made in view of the above-described circumstances, and is a list in which data portions having similar sounds are extracted from the sound data, and their position information in the sound data is grouped by similar data portions. It is an object to provide a similar voice list generation apparatus, a similar voice list generation method, and a similar voice list generation program.

The similar voice list generation device according to the present invention is
A comparison source voice block cutting means for sequentially cutting out a voice block of a predetermined time length from the voice data as a comparison source voice block;
A comparison target voice block cutting means for sequentially cutting out a voice block of a predetermined time length for comparison with the comparison source voice block as a comparison target voice block from the voice data;
A feature amount distance calculating means for calculating a feature amount of the comparison source speech block and a feature amount of the comparison destination speech block, and calculating a distance between both feature amounts;
Distance information storage means for storing distance information in which the distance calculated by the feature distance calculation means is associated with the positional information of the comparison source voice block and the comparison destination voice block;
After the distance is calculated for all the comparison source voice blocks and the comparison destination voice blocks, position information of voice blocks with similar voices is acquired based on the distance information stored in the distance information storage means. , A list generation means for generating a similar audio block list obtained by grouping these, and
Referring to the similar speech block list generated by the list generation means, extracts position information of a sound pattern in which a predetermined number or more of sound blocks belonging to any group continue, and extracts the extracted position information as a sound block Similar voice pattern list generation means for generating a similar voice pattern list grouped by voice patterns having the same configuration .

Further, the similar voice list generation method according to the present invention includes:
A comparison source audio block extraction step for sequentially extracting audio blocks of a predetermined time length from the audio data as comparison source audio blocks;
A comparison target voice block cutting step for sequentially cutting out a voice block of a predetermined time length for comparison with the comparison source voice block from the voice data as a comparison target voice block;
A feature amount distance calculating step of calculating a feature amount of the comparison source speech block and a feature amount of the comparison destination speech block, and calculating a distance between both feature amounts;
A distance information storage step of storing distance information in which the distance calculated in the feature amount distance calculation step and position information of the comparison source voice block and the comparison destination voice block are associated with each other in a predetermined storage unit;
After the distances are calculated for all the comparison source voice blocks and the comparison target voice blocks, the position information of the voice blocks with similar voices is acquired based on the distance information stored in the storage unit. A list generation step for generating a similar speech block list in which
With reference to the similar audio block list generated by the list generation step, the position information of the audio pattern in which a predetermined number or more of audio blocks belonging to any group are continuous is extracted, and the extracted position information is the audio block A similar voice pattern list generation step of generating a similar voice pattern list grouped by voice patterns having the same configuration.

Moreover, the similar sound list generation program according to the present invention is as follows.
Computer
A comparison source voice block cutting means for sequentially cutting out a voice block of a predetermined time length from the voice data as a comparison source voice block;
A comparison destination voice block cutting means for sequentially cutting out a voice block having a predetermined time length for comparison with the comparison source voice block from the voice data as a comparison destination voice block;
A feature distance calculation means for calculating a feature amount of the comparison source speech block and a feature amount of the comparison destination speech block and calculating a distance between both feature amounts;
Distance information storage means for storing distance information in which the distance calculated by the feature amount distance calculation means is associated with position information of the comparison source voice block and the comparison target voice block in a predetermined storage means;
After the distances are calculated for all the comparison source voice blocks and the comparison target voice blocks, the position information of the voice blocks with similar voices is acquired based on the distance information stored in the storage unit. A list generating means for generating a similar speech block list obtained by grouping
Referring to the similar speech block list generated by the list generation means, extracts position information of a sound pattern in which a predetermined number or more of sound blocks belonging to any group continue, and extracts the extracted position information as a sound block And a similar voice pattern list generation unit that generates a similar voice pattern list grouped by voice patterns having a common configuration .

  According to the present invention, since a location where similar audio content is recorded can be presented to the user in advance, user convenience is improved.

It is a block diagram which shows the structure of the similar audio | voice list production | generation apparatus which concerns on each embodiment of this invention. It is a flowchart which shows the procedure of a similar audio | voice list production | generation process. It is a figure for demonstrating extraction of the audio block of a comparison origin. It is a figure for demonstrating extraction of the audio | voice block of the first comparison destination. It is a figure for demonstrating the cutting of a flame | frame. It is a figure for demonstrating extraction of the audio | voice block of the next comparison destination. It is a figure which shows the example of a display of a similar audio | voice list | wrist. It is a flowchart which shows the procedure of a similar audio | voice pattern list production | generation process. It is a figure for demonstrating a similar audio | voice pattern.

  Hereinafter, an embodiment of a similar voice list generation device according to the present invention will be described with reference to the drawings.

(Embodiment 1)
As shown in FIG. 1, a similar audio list generation device 1 according to this embodiment includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, and an audio data storage. Unit 14, input unit 15, and display unit 16. These units are connected to each other via a system bus 17. The system bus 17 is a transmission path for transferring commands and data.

  The CPU 11 controls the ROM 12, the RAM 13, the audio data storage unit 14, the input unit 15 and the display unit 16 via the system bus 17, and accordingly, exchanges data with these.

  The ROM 12 stores an operation program (similar sound list generation program) for executing a similar sound list generation process to be described later, various data tables used in the similar sound list generation process, and the like. The RAM 13 is used as a working memory or the like in which data is temporarily saved when an operation program is executed.

  The audio data storage unit 14 is composed of, for example, a readable / writable nonvolatile semiconductor memory, a hard disk, or the like, and stores audio data recorded at a meeting or the like.

  The input unit 15 includes input devices such as a keyboard and a mouse, receives an operation input from a user, and sends the received signal (operation signal) to the CPU 11. The display unit 16 includes a display device such as a CRT or a liquid crystal monitor, and displays data such as characters and images supplied from the CPU 11. The display unit 16 functions as a first display unit and a second display unit under the control of the CPU 11.

  The CPU 11 operates according to the similar voice list generation program written in the ROM 12 to execute a similar voice list generation process as shown in the flowchart of FIG.

  The similar voice list generation process is started when the start instruction of this process is input by the user via the input unit 15 and the CPU 11 receives an operation signal to that effect. First, the CPU 11 cuts out a voice block (see voice block 1 in FIG. 3) having a predetermined time length (for example, 400 msec) from the head of the voice data stored in the voice data storage unit 14 (step S101).

  The CPU 11 cuts out the voice block immediately after the comparison source voice block from the voice data as the first counterpart (comparison destination) for comparison with the cut out comparison source voice block (step S102). As shown in FIG. 4, the time length of the comparison target audio block is ½ to 2 times the time length of the comparison source audio block (for example, 200 to 800 msec).

  The CPU 11 calculates the distance between the feature amount of the comparison source speech block and the feature amount of the comparison destination speech block (step S103). Here, the CPU 11 first cuts out a signal from both audio blocks with a predetermined time length (for example, 25 msec) and a predetermined shift width (for example, 10 msec), and uses this as a frame (see FIG. 5).

  Next, the CPU 11 compares the frame of the comparison source voice block and the frame of the comparison destination voice block by DP (Dynamic Programming) matching, and determines the shortest distance between the comparison source voice block and the comparison destination voice block. Distance.

In the calculation of the distance between frames, an MFCC (Mel Frequency Cepstrum Coefficient) is calculated from an audio sample of the frame, and the distance between vectors of the MFCCs is set as a distance between frames. For example, an angle between vectors is used as the distance between vectors. The angle between the vectors of N vectors {x _n | n = 1..N} and {y _n | n = 1..N} is obtained by the following Equation 1.

  As described above, when the distance between the feature amount of the comparison source speech block and the feature amount of the comparison destination speech block is calculated, the CPU 11 calculates the calculated distance and the position information of both speech blocks ( Information (distance information) that associates the information on the reproduction position in the audio data) with the RAM 13 is temporarily stored. If the comparison target audio block is not final (NO in step S104), the CPU 11 cuts out the next comparison target audio block from the audio data. At this time, as shown in FIG. 6, the CPU 11 cuts out the next comparison target audio block from the position shifted by a predetermined shift width (for example, 50 msec) (step S105).

  In the same manner as described above, the distance between the feature amount of the comparison source speech block and the feature amount of the comparison destination speech block is calculated (step S103).

  On the other hand, if the comparison target audio block is final (YES in step S104), the CPU 11 determines that the comparison source audio block is not the final (NO in step S106), and the next comparison source audio block from the audio data. Is cut out (step S107). At this time, as shown in FIG. 3, the CPU 11 cuts out the next comparison source audio block from the position shifted by a predetermined shift width (for example, 50 msec).

  Then, the CPU repeats the processing from step 102.

  If it is determined in step S106 that the comparison source audio block is final (YES in step S106), the CPU 11 selects a similar one from all the above audio blocks based on all the distance information stored in the RAM 13. The positional information of the related audio blocks is acquired, and a grouped data list is generated (step S108). Specifically, the CPU 11 extracts a predetermined number from the shorter distances among the calculated distances between the feature amounts. Then, in order from the shortest distance, the position information of the pair (comparison source and comparison target speech block) is stored as one group in a table developed in the RAM 13. At this time, each pair is not classified into a different group, and when one of the pair to be grouped corresponds to one of the previously grouped pairs, the pair to be grouped is It shall be classified into the same group.

  In this way, a data list (similar audio block list) in which similar audio blocks are grouped is generated.

  The user can confirm the contents of the similar audio block list by printing or displaying the similar audio block list in a predetermined manner. For example, as shown in FIG. 7, the CPU 11 may color-similar similar audio blocks for each group and display them on the display unit 16 (in FIG. 7, the difference in color is represented by the difference in the fill pattern. ing.). In this case, if a predetermined color is assigned in the order of the number of pairs belonging to the group, the user can easily visually recognize the position of the frequently occurring audio block in the audio data in which the contents of the conference or the like are recorded. .

  In addition, when the contents of the similar audio block list are represented by characters or the like, the positions of frequently occurring audio blocks are highly visible to the user as described above by listing them in the order of the number of pairs belonging to the group. Can be presented.

  Generally, in a conference or the like, voice contents (words) frequently uttered by a speaker are often related to important points in the conference or the like.

  As described above, the similar voice list generation device 1 of the present embodiment can present to the user in advance a location where frequently-occurring voice is recorded, so that the user picks up and plays back that location and partially records it. It is possible to grasp important points just by listening to the contents. Therefore, user convenience is significantly improved.

(Embodiment 2)
Next, the similar voice list generation device according to the second embodiment will be described. The configuration of the similar voice list generation device of the present embodiment is the same as that of the similar voice list generation device 1 of the first embodiment (see FIG. 1). The similar voice list generation apparatus of the present embodiment is similar to the similar voice list generation apparatus 1 of the first embodiment in that a similar voice pattern list is further generated based on the generated similar voice block list in the similar voice list generation process. Different.

  In the similar speech block list, speech blocks with high similarity are grouped, but there are speech blocks whose length is less than the morpheme, and the actual speech content (word) may be different even in the same group. obtain. Therefore, the similar sound list generation apparatus of the present embodiment extracts a block (speech pattern) of a plurality of sound blocks after the generation of the similar sound block list in the similar sound list generation process, and the sound patterns are similar to each other. A process (similar voice pattern list generation process) for distributing the current ones to the same group is executed.

  FIG. 8 is a flowchart showing the procedure of the similar voice pattern list generation process incorporated in the similar voice list generation process (see FIG. 2). The CPU 11 refers to the generated similar speech block list and extracts position information about a block (speech pattern) of speech blocks continuous for a predetermined number or more (step S201).

  Next, the CPU 11 generates a data list (similar voice pattern list) in which the extracted voice pattern position information is grouped with the same voice block configuration (step S202). In the example of FIG. 9, the audio patterns composed of the audio blocks (5, 1, 4, 6, 3, 2) are classified into the same group. Here, each numerical value indicating an audio block represents a group number to which the audio block belongs.

  The user can confirm the contents of the similar voice pattern list by printing or displaying the similar voice pattern list in a predetermined manner. In this case, by displaying the list in order from the largest number of voice patterns belonging to the group, or by assigning and displaying a predetermined color, the user frequently appears in the voice data in which the content of the meeting or the like is recorded. The position of the voice content (word) to be played can be easily recognized.

  In the similar sound pattern list, similar grouping is performed in units of sound blocks (sound patterns), and therefore, sound patterns belonging to the same group are highly likely to have the same sound content. Therefore, the user can more surely grasp important points in a meeting or the like only by picking up and reproducing the recording position of a frequently occurring voice pattern and only partially listening to the recorded content.

  The present invention is not limited to the above-described embodiments, and various modifications can be made without departing from the scope of the present invention.

  For example, in the similar sound list generation process, a process of excluding position information of a sound block with a small number of similarities (for example, not reaching a preset number) from the similar sound block list may be added. Similarly, a process of excluding position information of a voice pattern with a small number of similarities from the similar voice pattern list may be added.

  In the first embodiment, in the generation of the similar audio block list, a predetermined number is extracted in order from the shorter distance between the calculated feature amounts as a pair of similar audio blocks. However, the present invention is not limited to this. You may make it extract to 1% of all the things shorter than a predetermined threshold value, or the order with a short distance.

  Further, not only the MFCC but also an LSP (Line Spectrum Pair) coefficient or FFT (fast Fourier series expansion) may be adopted as the feature amount.

  In addition, Manhattan distance (see Equation 2), Euclidean distance (see Equation 3), Canberra distance (see Equation 4), or the like may be used for calculating the distance between vectors.

  Also, by applying the program, it is possible to cause an existing personal computer (PC) or the like to function as the similar voice list generation device according to the present invention. That is, the program as executed by the CPU 11 described above is applied to an existing PC or the like, and a control device (CPU or the like) such as the PC executes the program, whereby the similar voice list generation device according to the present invention is obtained. It is possible to make it work.

  Such a program distribution method is arbitrary. For example, the program can be distributed by being stored in a recording medium such as a memory card, or can be distributed through a communication network such as the Internet.

  DESCRIPTION OF SYMBOLS 1 ... Similar audio | voice list production | generation apparatus, 11 ... CPU, 12 ... ROM, 13 ... RAM, 14 ... Audio | voice data storage part, 15 ... Input part, 16 ... Display part, 17 ... System bus

Claims (6)

A comparison source voice block cutting means for sequentially cutting out a voice block of a predetermined time length from the voice data as a comparison source voice block;
A comparison target voice block cutting means for sequentially cutting out a voice block of a predetermined time length for comparison with the comparison source voice block as a comparison target voice block from the voice data;
A feature amount distance calculating means for calculating a feature amount of the comparison source speech block and a feature amount of the comparison destination speech block, and calculating a distance between both feature amounts;
Distance information storage means for storing distance information in which the distance calculated by the feature distance calculation means is associated with the positional information of the comparison source voice block and the comparison destination voice block;
After the distance is calculated for all the comparison source voice blocks and the comparison destination voice blocks, position information of voice blocks with similar voices is acquired based on the distance information stored in the distance information storage means. , A list generation means for generating a similar audio block list obtained by grouping these, and
Referring to the similar speech block list generated by the list generation means, extracts position information of a sound pattern in which a predetermined number or more of sound blocks belonging to any group continue, and extracts the extracted position information as a sound block Similar voice pattern list generation means for generating a similar voice pattern list grouped by voice patterns having the same configuration of
A similar voice list generation device characterized by the above. And further comprising display means for displaying the contents of the similar voice pattern list generated by the similar voice pattern list generation means .
The similar speech list generation device according to claim 1 , wherein The display means displays the position information of the sound pattern in a different manner for each group.
The similar voice list generation device according to claim 2 , wherein: The display means displays the position information of the voice pattern having a large number of similarities in a manner that the user can identify.
Similar voice list generating apparatus according to claim 2 or 3, characterized in that. A comparison source audio block extraction step for sequentially extracting audio blocks of a predetermined time length from the audio data as comparison source audio blocks;
A comparison target voice block cutting step for sequentially cutting out a voice block of a predetermined time length for comparison with the comparison source voice block from the voice data as a comparison target voice block;
A feature amount distance calculating step of calculating a feature amount of the comparison source speech block and a feature amount of the comparison destination speech block, and calculating a distance between both feature amounts;
A distance information storage step of storing distance information in which the distance calculated in the feature amount distance calculation step and position information of the comparison source voice block and the comparison destination voice block are associated with each other in a predetermined storage unit;
After the distances are calculated for all the comparison source voice blocks and the comparison target voice blocks, the position information of the voice blocks with similar voices is acquired based on the distance information stored in the storage unit. A list generation step for generating a similar speech block list in which
With reference to the similar audio block list generated by the list generation step, the position information of the audio pattern in which a predetermined number or more of audio blocks belonging to any group are continuous is extracted, and the extracted position information is the audio block A similar voice pattern list generation step for generating a similar voice pattern list grouped by a voice pattern having a common configuration of
A similar voice list generation method characterized by the above. Computer
A comparison source voice block cutting means for sequentially cutting out a voice block of a predetermined time length from the voice data as a comparison source voice block;
A comparison destination voice block cutting means for sequentially cutting out a voice block having a predetermined time length for comparison with the comparison source voice block from the voice data as a comparison destination voice block;
A feature distance calculation means for calculating a feature amount of the comparison source speech block and a feature amount of the comparison destination speech block and calculating a distance between both feature amounts;
Distance information storage means for storing distance information in which the distance calculated by the feature amount distance calculation means is associated with position information of the comparison source voice block and the comparison target voice block in a predetermined storage means;
After the distances are calculated for all the comparison source voice blocks and the comparison target voice blocks, the position information of the voice blocks with similar voices is acquired based on the distance information stored in the storage unit. A list generating means for generating a similar speech block list obtained by grouping
Referring to the similar speech block list generated by the list generation means, extracts position information of a sound pattern in which a predetermined number or more of sound blocks belonging to any group continue, and extracts the extracted position information as a sound block A similar voice pattern list generating means for generating a similar voice pattern list grouped by a voice pattern having a common configuration of
A similar speech list generation program characterized by the above.

Images