Sign language word duration calculation device and program, and sign language CG image generation device and program

Abstract

To provide a sign language word time length calculation device capable of calculating a time length of a sign language word corresponding to a sign language motion.SOLUTION: A sign language word time length calculation device 1 includes: a word frequency vector generation part 22 that generates a word frequency vector, in which the number of whole sign language words is element number, and frequency of each sign language word appearing in a sign language word string is element; and a multiple regression analysis part 23 which performs multiple regression analysis on a multivariable linear model expressing the time length of the sign language word string by inner product of the word frequency vector and the word time length vector with the word time length of each sign language word as an element, using a database pairing a sign language word string with a time length of a sign language motion of the sign language word string, and calculates time length of the sign language word.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a sign language word duration calculation device and a program therefor, and a sign language CG generation device and a program therefor. [Background technology]

[0002] Traditionally, sign language services such as sign language interpretation and sign language broadcasting have been commonly provided for the hearing impaired. In recent years, in order to expand sign language services, research has been progressing on sign language animation using sign language CG (computer graphics). Sign language CG is basically generated by translating text into a string of sign language words and concatenating sign language word CGs that correspond to the individual words in the translated string of sign language words. As a method for translating text into a string of sign language words, for example, a machine translation device technology has been disclosed that generates a bilingual pair by transcribing Japanese text and a string of sign language words from broadcast footage (such as sign language news), and then uses the bilingual pair to translate any Japanese text into a string of sign language words (see Patent Document 1). In addition, as a method for connecting sign language word CG, a technology has been disclosed in which sign language motions of sign language words converted (translated) from Japanese text are read out in order from a sign language motion database and connected to generate sign language CG (see Patent Document 2). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2013-186673 [Patent Document 2] Japanese Patent Application Publication No. 2017-151757 Summary of the Invention [Problem to be solved by the invention]

[0004] Conventional sign language motions are generated for each sign language word. Furthermore, sign language CG generated by conventional methods simply connects sign language word CG for each word in a translated sign language word string. In other words, conventional sign language CG only takes into account the order of sign language words, and does not take into account the playback time for each word during sign language motion. Therefore, when sign language CG generated using conventional methods is played back, each sign language word CG is played for a length of time that is unrelated to the series of actions, resulting in a lack of naturalness. As such, sign language CG generated using conventional methods lacks naturalness, making it difficult for people with hearing impairments to understand the sign language content.

[0005] The present invention has been made in consideration of such problems, and aims to provide a sign language word duration calculation device and a program therefor that are capable of calculating the duration of individual sign language words in a series of sign language actions, as well as a sign language CG image generation device and a program therefor that are capable of generating sign language CG images corresponding to the duration of each sign language word. [Means for solving the problem]

[0006] In order to solve the above problem, the sign language word duration calculation device of the present invention is a sign language word duration calculation device that calculates the duration of the sign language movements of each sign language word from a database that pairs sign language word strings with the durations of the sign language movements of the sign language word strings, and is configured to include a word frequency vector generation unit and a multiple regression analysis unit.

[0007] In such a configuration, the sign language word duration calculation device generates, by the word frequency vector generation unit, a word frequency vector whose elements are the total number of sign language words and whose elements are the frequency of each sign language word appearing in the sign language word string.

[0008] Then, the sign language word duration calculation device uses a database to perform multiple regression analysis on a multivariate linear model that represents the duration of a sign language word sequence using the inner product of a word frequency vector and a word duration vector whose elements are the word durations of each sign language word, thereby calculating the duration of each sign language word. The sign language word duration calculation device can be operated by a sign language word duration calculation program that causes a computer to function as each of the above-mentioned units.

[0009] In addition, in order to solve the above problem, the sign language CG image generation device of the present invention is a sign language CG image generation device that generates sign language CG images from a string of sign language words using the duration of the sign language movements of each sign language word calculated by multiple regression analysis from a database that pairs a string of sign language words with the duration of the sign language movements of the string of sign language words, and is configured to include a sign language word CG image generation unit, a duration adjustment unit, and an image connection unit.

[0010] In this configuration, the sign language CG image generation device uses the sign language word CG image generation unit to generate, for each sign language word constituting the input sign language word string, a sign language word CG image that expresses the sign language action in CG using motion data corresponding to the sign language word, thereby generating a CG image for each sign language word.

[0011] The sign language CG image generation device then uses a duration adjustment unit to adjust the CG images of sign language words to the durations of the sign language actions of the sign language words calculated by multiple regression analysis, thereby making the durations of the CG images of each sign language word equal to the durations of the sign language actions of each sign language word in the sign language actions of the sequence of sign language words. The sign language CG image generation device then generates a sign language CG image by connecting the time length adjusted sign language word CG images in the order of the sign language word string using the image connection unit. The sign language CG image generation device can be operated by a sign language CG image generation program that causes a computer to function as each of the above-mentioned units. [Effects of the Invention]

[0012] According to the present invention, the duration of the sign language motion of each sign language word can be calculated from the duration of the sign language motion of a sequence of multiple sign language words. Also, according to the present invention, a sign language CG image can be generated from a sequence of sign language words, in which the motion duration of each sign language word is a predetermined duration. As a result, the present invention can generate a sign language CG image that naturally expresses each sign language action in a series of sign language actions, with the same duration as the actual sign language actions. [Brief explanation of the drawings]

[0013] [Figure 1] 1 is a block diagram showing the configuration of a sign language word duration calculation device according to a first embodiment of the present invention. [Figure 2] 4 is an explanatory diagram for explaining the contents stored in a word string duration storage unit; FIG. [Figure 3] FIG. 2 is an explanatory diagram for explaining the contents stored in a word duration storage unit; [Figure 4] FIG. 2 is an explanatory diagram illustrating the contents of a word frequency vector generated by a word frequency vector generation unit. [Figure 5] 10 is an explanatory diagram for explaining the contents of a word duration vector generated by a word string duration modeling unit. FIG. [Figure 6] 3 is a flowchart showing the operation of the sign language word duration calculation device according to the first embodiment of the present invention. [Figure 7] FIG. 10 is a block diagram showing the configuration of a sign language CG image generation device according to a second embodiment of the present invention. [Figure 8] 10 is a flowchart showing the operation of the sign language CG image generation device according to the second embodiment of the present invention. [Figure 9] FIG. 2 is an explanatory diagram for schematically explaining a sign language CG image generated by a sign language CG image generation device. DETAILED DESCRIPTION OF THE INVENTION

[0014] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. First Embodiment [Configuration of a sign language word duration calculation device] The configuration of a sign language word duration calculation device 1 according to a first embodiment of the present invention will be described with reference to FIG.

[0015] The sign language word duration calculation device 1 calculates the duration of the sign language action of each sign language word from a database that pairs sign language word strings with the duration of the sign language action of the sign language word string. The sign language word duration calculation device 1 includes a storage unit 10 and a control unit 20.

[0016] The storage unit 10 is a general storage medium such as a hard disk or semiconductor memory. The storage unit 10 includes a word string duration storage unit 11 and a word duration storage unit 12 . The word string duration storage unit 11 and the word duration storage unit 12 may be configured to be stored in separate areas within the same storage medium, or may be configured to be stored in different storage media.

[0017] The word string duration storage unit 11 stores a plurality of Japanese texts TX as a corpus, a sign language word string WS obtained by translating the Japanese text TX, and a time length of a sign language action corresponding to the sign language word string WS (word string duration T WS ) and are stored in advance in association with each other. The sign language word sequence WS can be generated by transcribing the sign language words that appear in the video of the sign language news, etc., corresponding to the Japanese text TX. WS can use the actual duration of the sign language actions in the video of the sign language word sequence WS corresponding to the Japanese text TX, along with the transcription of the sign language words.

[0018] For example, as shown in Figure 2, if a certain Japanese text TX is "The amount of snow that will fall by tonight is in all places where it is heavy...", the corresponding sign language word string WS would be, for example, "N, dark, until, snow, amount, N, a lot, place,...". Note that "N" is a special symbol that indicates the action of nodding, and is treated as one word. The time taken for all the sign words W (N, dark, until, ...) that make up the sign word sequence WS to be expressed in sign language is the word sequence duration T WS Let's say.

[0019] For ease of understanding, the word string duration storage unit 11 stores Japanese text TX, sign language word strings WS, and word string durations T WS However, in the present invention, Japanese text TX is not essential. That is, the word string duration storage unit 11 stores at least the sign language word string WS and the word string duration T WS It is sufficient that the above are stored in association with each other. Furthermore, the sign language words that make up the sign language word string WS may be the sign language vocabulary itself, or may be identifiers that identify the vocabulary. The number of sign language word sequences WS, i.e., the number of Japanese texts TX, is approximately tens of thousands. It is assumed that the entire sign language vocabulary is covered by all sign language word sequences WS. The number of sign language words is approximately 10,000.

[0020] The word duration storage unit 12 stores a sign language word W and a duration of a sign language action corresponding to one sign language word W (word duration T W ) is stored. The word duration storage unit 12 stores sign language words W corresponding to the number of vocabulary words and word duration T W The following is stored. The sign language word W may be a sign language word itself, or may be an identifier that identifies the word.

[0021] Word duration T W are parameter values ​​to be optimized by the multiple regression analysis unit 23. Also, the word duration T W is initialized by the word duration initialization unit 21 and updated by the multiple regression analysis unit 23. In addition, the word duration T W is referred to by the multiple regression analysis unit 23. Here, the word duration T of sign language word W is WIt is assumed that there is no variation in all sign language word strings WS stored in the word string duration storage unit 11. In other words, it is assumed that one sign language word W operates with a specific duration.

[0022] The control unit 20 controls the entire sign language word duration calculation device 1. The control unit 20 loads a program (sign language word duration calculation program) stored in a hard disk, ROM, etc. into memory, and the computer (CPU) reads the program and causes it to function as each unit described below. The control unit 20 includes a word duration initialization unit 21, a word frequency vector generation unit 22, and a multiple regression analysis unit 23.

[0023] The word duration initialization unit 21 initializes the duration of a sign language word to be optimized (word duration). The word duration initialization unit 21 initializes the word duration T for each sign language word W stored in the word duration storage unit 12. W Initialize. As shown in FIG. 3, the word duration initialization unit 21 initializes sign language words W1, W2, . . . , W n For (n: number of words), the initial value of word duration t 1_0 ,t 2_0 ,…,t n_0 Set.

[0024] The initial value may be any value, such as a random value. However, in order to optimize the time length, it is preferable to use an approximately appropriate time as the action time of a sign language word as the initial value. The word duration initialization unit 21 stores the initial value of the word duration for each sign language word W in the word duration storage unit 12 . The word duration initialization unit 21 operates only once at startup.

[0025] The word frequency vector generating unit 22 generates a word frequency vector for each sign language word string WS, with the number of elements being the total number of sign language words and the elements being the frequency of each sign language word appearing in the sign language word string WS. Here, the word frequency vector generation unit 22 generates a vector (word frequency vector v) shown in the following equation (1).

[0026]

number

[0027] c1,c2,…,c n (n: number of words) is the number of sign language words W1, W2, ..., W included in the sign language word sequence WS. n Indicates the number of. For example, when the sign language word string WS is the one illustrated in FIG. 2, the word frequency vector generation unit 22 sets the number of sign language words included in the sign language word string WS to the same element position as the sign language words, as shown in FIG. 4, and generates a word frequency vector v shown in the following equation (2).

[0028]

number

[0029] In this case, the word frequency vector v represents 0 occurrences of the sign language word "bright," 1 occurrence of "dark," ..., and 2 occurrences of "N." The word frequency vector generation unit 22 outputs the generated word frequency vector v to the multiple regression analysis unit 23.

[0030] The multiple regression analysis unit 23 performs multiple regression analysis on a multivariate linear model that represents the duration of a sequence of sign language words using the inner product of a word frequency vector and a word duration vector whose elements are the word durations of each sign language word, and calculates the duration of each sign language word.

[0031] The multiple regression analysis unit 23 performs multiple regression analysis on a multivariable linear model in which the inner product of a word duration vector, whose elements are the number of vocabulary words in the entire sign language and whose elements are explanatory variables that are parameters for the word duration of sign language words, and a word frequency vector, is equal to the objective variable. Here, the multiple regression analysis unit 23 calculates the sign language words W1, W2, . . . , W as shown in FIG. nThe number of elements is the number of vocabulary words n, and the explanatory variables are the time lengths of sign language words, t1, t2, t3, ..., t n A word duration vector t is generated as shown in the following equation (3) with the elements:

[0032]

number

[0033] The multiple regression analysis unit 23 can predict the sign language action duration using the multivariate linear model shown in the following equation (4) by taking the inner product of the word frequency vector v generated by the word frequency vector generation unit 22 and the word duration vector t, where T represents the transpose of the vector.

[0034]

number

[0035] The multiple regression analysis unit 23 performs multiple regression analysis using a database (word sequence duration storage unit 11) that pairs sign language word sequences with the durations of sign language actions of the sign language word sequences. This allows the multiple regression analysis unit 23 to calculate the durations of the sign language actions of sign language words, which are explanatory variables of the multivariate linear model.

[0036] The multiple regression analysis in the multiple regression analysis unit 23 may be performed using a general method. For example, the multiple regression analysis unit 23 can use the least squares method as the multiple regression analysis. In this case, the multiple regression analysis unit 23 calculates the predicted time length (T pred ) and the word string duration (T true ) to minimize the error (sum of squared residuals). Since the number of explanatory variables is enormous and complicated, the gradient descent method may be used for the multiple regression analysis. That is, the multiple regression analysis unit 23 can calculate the components of t by minimizing the evaluation function E shown in the following equation (5) with the error being the mean square error.

[0037]

number

[0038] where t i are the parameters t1, t2, t3, ..., t at the i-th update n Ns indicates the number of sign language word strings WS (number of sentences) stored in the word string duration storage unit 11. T true (s) is the word sequence duration T of the sth sign language word sequence WS. WS (Correct answer time length T true ) indicates T pred (s) is the predicted duration T of the sth sign language word sequence WS. pred Shows. These parameters t1,t2,t3,…,t n can be updated using the following equation (6).

[0039]

number

[0040] Here, α is a learning rate (hyperparameter) that indicates how much the parameters are updated according to the gradient.

[0041] With the above configuration, the sign language word duration calculation device 1 calculates the word string duration T WS From this, the duration of the sign language word corresponding to the sign language action can be calculated.

[0042] [Operation of the sign language word duration calculation device] Next, the operation of the sign language word duration calculation device 1 will be described with reference to FIG. 6 (for the configuration, refer to FIG. 1 as appropriate). The word string duration storage unit 11 stores at least a sign language word string WS and the duration of a sign language action corresponding to the sign language word string WS (word string duration T WS ) are stored in advance in association with each other. Also, it is assumed that the word duration storage unit 12 stores in advance sign language words W corresponding to the number of vocabulary words.

[0043] In step S1, the word duration initialization unit 21 initializes the word duration T for each sign language word W stored in the word duration storage unit 12. W Initialize. In step S2, the word frequency vector generation unit 22 generates, for each sign language word sequence WS, a word frequency vector v (see equation (1)) whose elements are the vocabulary count of sign language words and whose elements are the occurrence frequencies of sign language words included in the sign language word sequence WS.

[0044] In step S3, the multiple regression analysis unit 23 calculates the inner product of the word duration vector t (see equation (3)) whose elements are parameters of the durations of the sign language vocabulary words and the word frequency vector v generated in step S2 as the predicted duration T of the sign language word string WS. pred A multivariate linear model (see equation (4)) is subjected to multiple regression analysis to calculate the duration of the sign language action for the sign language word. By the above operation, the sign language word duration calculation device 1 calculates the word string duration T WS From the above, the time length of the sign language word corresponding to the sign language action (word time length T W ) can be calculated.

[0045] Second Embodiment [Configuration of the sign language CG image generation device] Next, with reference to FIG. 7, the configuration of a sign language CG image generation device 3 according to a second embodiment of the present invention will be described.

[0046] The sign language CG image generating device 3 generates a sign language CG image from a string of sign language words using the word duration of each sign language word calculated by the sign language word duration calculating device 1 (see FIG. 1). In other words, the sign language CG image generation device 3 generates a sign language CG image from a string of sign language words using the duration of the sign language movements of the sign language words calculated by multiple regression analysis from a database that pairs strings of sign language words with the duration of the sign language movements of the string of sign language words. The sign language CG image generation device 3 includes a storage unit 30 and a control unit 40.

[0047] The storage unit 30 is a general storage medium such as a hard disk or semiconductor memory. The storage unit 30 includes a motion data storage unit 31 and a word duration storage unit 32 . The motion data storage unit 31 and the word duration storage unit 32 may be configured to be stored in separate areas within the same storage medium, or may be configured to be stored in different storage media.

[0048] The motion data storage unit 31 is a database in which motion data D that defines sign language actions (sign language motions) in association with sign language words W is stored in advance. The motion data D is data describing the joint movements in sign language actions. For example, the motion data D is data listing a skeletal structure (bone model) with joints and the joint angles in each frame, and is motion capture data described in the BVH (Biovision Hierarchy) format. The motion data storage unit 31 is referenced by the sign language word CG image generation unit 41.

[0049] The word duration storage unit 32 stores a sign language word W and a time length (word duration) T of a sign language action corresponding to one sign language word W. W The word duration T corresponding to the sign language word W is stored in advance. W uses the duration calculated by the sign language word duration calculation device 1 described with reference to FIG. The word duration storage unit 32 is referenced by the duration adjustment unit 42 .

[0050] The sign language words W in the motion data storage unit 31 and the word duration storage unit 32 may be the sign language words themselves, or may be identifiers that identify the words. The motion data storage unit 31 and the word duration storage unit 32 store motion data D and word duration T for a sign language word W. W It may be a single database that associates the above.

[0051] The control unit 40 controls the entire sign language CG image generation device 3. The control unit 40 loads a program (sign language CG image generation program) stored in a hard disk, ROM, or the like into memory, and the computer (CPU) reads the program and causes it to function as each unit described below. The control unit 40 includes a sign language word CG image generation unit 41, a time length adjustment unit 42, and an image connection unit 43.

[0052] The sign language word CG image generating unit 41 generates a sign language word CG image I in which the sign language action is expressed by CG for each sign language word constituting the input sign language word string WS. W This generates: The sign language word CG image generation unit 41 reads out the motion data D corresponding to the sign language word W constituting the sign language word string WS from the motion data storage unit 31. Then, the sign language word CG image generation unit 41 generates a CG image (sign language word CG image I) corresponding to the joint movements described in the motion data D. W ) The sign language word CG image generating unit 41 generates the sign language word CG image I W is output to the time length adjustment unit 42 together with the sign language word.

[0053] The time length adjusting unit 42 adjusts the sign language word CG image I generated by the sign language word CG image generating unit 41. W The time length of the sign language word is adjusted to be the word time length corresponding to the sign language word. Here, the time length adjustment unit 42 adjusts the sign language word CG image I W The word duration T of the corresponding sign language word W W is read from the word duration storage unit 32, and the sign language word CG image IW The time length of the word T W Adjust to match.

[0054] That is, the time length adjusting unit 42 adjusts the sign language word CG image I W The time duration of the corresponding sign language word W is the word duration T W If the time difference is shorter than the CG image I, the number of frames corresponding to the difference in time length is W By inserting it evenly, the time length is increased. For example, CG video of sign language words I W When the number of frames of the sign language word CG image I is M frames and the number of frames corresponding to the difference in time length is m frames, the time length adjustment unit 42 W The last frame is copied and inserted after the last frame for every M / m frame video section.

[0055] The time length adjusting unit 42 also adjusts the sign language word CG image I W The time duration of the corresponding sign language word W is the word duration T W If the time difference is longer than the CG image I, the number of frames corresponding to the difference in time length is W By removing evenly from the For example, if the number of frames in the sign language word CG image is M frames and the number of frames corresponding to the difference in time length is m frames, the time length adjustment unit 42 adjusts the sign language word CG image I W An arbitrary frame (for example, the last frame) is deleted for each video section of M / m frames.

[0056] As a result, the time length adjusting unit 42 adjusts the sign language word CG image I W The time length of the sign language word W is stored in the word time length storage unit 32. W can be adjusted to. The time length adjusting unit 42 adjusts the time length of the sign language word CG image I W is output to the video connection unit 43.

[0057] The video linking unit 43 links the sign language word CG video I whose time length has been adjusted by the time length adjusting unit 42. Ware concatenated in the order of the sign language word string. The image connection unit 43 connects the sign language word CG images I to the sign language word string WS. W By connecting these, a sign language CG image I corresponding to the sign language word sequence WS is created. WS Generate. With the above configuration, the sign language CG image generation device 3 can generate, from a string of sign language words, a sign language CG image with movements that are close to the rhythm of actual sign language.

[0058] [Operation of the sign language CG image generation device] Next, the operation of the sign language CG image generation device 3 will be described with reference to FIG. 8 (for the configuration, see FIG. 7 as appropriate). It is assumed that the motion data storage unit 31 stores motion data D in advance in association with the sign language word W. The word duration storage unit 32 stores word durations T W is assumed to be stored.

[0059] In step S10, the sign language word CG image generation unit 41 generates a sign language word CG image I in which the sign language action is expressed in CG using the motion data D stored in the motion data storage unit 31 for each sign language word W constituting the input sign language word string WS. W Generate. In step S11, the time length adjusting unit 42 adjusts the sign language word CG image I generated in step S10. W Adjust the time length. Here, the time length adjustment unit 42 adjusts the sign language word CG image I W The time length of the sign language word W is calculated by dividing the time length of the sign language word W by the time length T W Adjust to match.

[0060] In step S12, the duration adjustment unit 42 determines whether or not the duration adjustment has been completed for all sign language words in the input sign language word string WS. Here, for all sign language words, there are CG images of sign language words. WIf the generation of the sign language word and the adjustment of its duration have not been performed (No in step S12), the duration adjustment unit 42 returns the operation to step S10 and continues the process for the next sign language word.

[0061] On the other hand, for all sign language words, CG image I W When the generation and time length adjustment are performed (Yes in step S12), the video connection unit 43 generates all the sign language word CG videos I whose time lengths have been adjusted in step S11. W Concatenate the following. As a result, the image connection unit 43 generates a sign language CG image corresponding to the sign language word string. Through the above operations, the sign language CG image generation device 3 can generate sign language CG images in which the duration of the sign language word CG images has been adjusted by the duration of the sign language word calculated by the sign language word duration calculation device 1.

[0062] For example, when the sign language word string WS "N, dark, until, snow, amount, N, a lot, place, ..." shown in FIG. 2 is input, the sign language CG image generation device 3 generates a predetermined word duration T W Sign Language Word CG Video I W By generating and connecting these, the sign language CG image I WS As a result, the sign language CG image generation device 3 generates the sign language CG image I according to the duration of each sign language action, for example, the sign language action for "N" is 0.32 seconds and the sign language action for "dark" is 0.58 seconds. WS can be generated. This allows the sign language CG image generation device 3 to generate sign language CG images that are close to the rhythm of actual sign language. Although the embodiments of the present invention have been described above, the present invention is not limited to these embodiments. For example, the sign language word duration calculation device 1 and the sign language CG image generation device 3 are configured to include the internal storage units 10 and 30, respectively, but the storage units 10 and 30 may be separate external storage devices. [Explanation of symbols]

[0063] 1. Sign language word duration calculation device 10 Storage section 11 Word sequence duration memory section (database) 12 Word duration memory section 20 Control Unit 21 Word duration initialization unit 22 Word frequency vector generation unit 23 Multiple Regression Analysis Section 3. Sign language CG image generation device 30 Storage section 31 Motion data storage unit 32 Word duration memory section 40 Control Unit 41 Sign language word CG image generation unit 42 Time length adjustment section 43 Video connection section

Claims

1. A sign language word duration calculation device that calculates the duration of a sign language action of each sign language word from a database that pairs a sign language word string with the duration of a sign language action of the sign language word string, comprising: a word frequency vector generation unit that generates a word frequency vector in which the number of elements is the total number of sign language words and the frequency of each sign language word appearing in the sign language word string is an element; a multiple regression analysis unit that uses the database to perform multiple regression analysis on a multivariate linear model that represents the duration of the sign language word sequence by using an inner product of the word frequency vector and a word duration vector whose elements are the word durations of each sign language word, and calculates the durations of the sign language words; A sign language word duration calculation device comprising:

2. 2. The sign language word duration calculation device according to claim 1, wherein the multiple regression analysis unit calculates the duration of the sign language word by minimizing an error between the predicted duration of the sign language action and the duration of the sign language word sequence stored in the database in the multivariate linear model.

3. A sign language word duration calculation program for causing a computer to function as the sign language word duration calculation device according to claim 1 or 2.

4. A sign language CG image generation device that generates a sign language CG image from a sign language word sequence using durations of sign language actions for each sign language word calculated by multiple regression analysis from a database that pairs a sign language word sequence with durations of sign language actions for the sign language word sequence, comprising: a sign language word CG image generation unit that generates, for each sign language word constituting the input sign language word string, a sign language word CG image that expresses a sign language action using CG, using motion data corresponding to the sign language word; a duration adjustment unit that adjusts the CG image of the sign language word so that the duration of the sign language action of the sign language word calculated by the multiple regression analysis is equal to the duration of the sign language action of the sign language word; an image connection unit that connects the time-length-adjusted sign language word CG images in the order of the sign language word string to generate the sign language CG image; A sign language CG image generation device comprising:

5. The sign language CG image generation device of claim 4, characterized in that when the time length of the sign language word CG image is shorter than the time length of the corresponding sign language word, the time length adjustment unit evenly inserts a number of frames equivalent to the difference in time length into the sign language word CG image, and when the time length of the sign language word CG image is longer than the time length of the corresponding sign language word, the time length adjustment unit evenly deletes a number of frames equivalent to the difference in time length from the sign language word CG image.

6. A sign language CG image generation program for causing a computer to function as the sign language CG image generation device according to claim 4 or 5.

Images