+

WO2018122919A1 - Dispositif de recherche basée sur un mot de sentiment - Google Patents

Dispositif de recherche basée sur un mot de sentiment Download PDF

Info

Publication number
WO2018122919A1
WO2018122919A1 PCT/JP2016/088704 JP2016088704W WO2018122919A1 WO 2018122919 A1 WO2018122919 A1 WO 2018122919A1 JP 2016088704 W JP2016088704 W JP 2016088704W WO 2018122919 A1 WO2018122919 A1 WO 2018122919A1
Authority
WO
WIPO (PCT)
Prior art keywords
sensitivity
vector
unit
information
proper name
Prior art date
Application number
PCT/JP2016/088704
Other languages
English (en)
Japanese (ja)
Inventor
貴弘 大塚
咲子 二本柳
相川　勇之
Original Assignee
三菱電機株式会社
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 三菱電機株式会社 filed Critical 三菱電機株式会社
Priority to PCT/JP2016/088704 priority Critical patent/WO2018122919A1/fr
Publication of WO2018122919A1 publication Critical patent/WO2018122919A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor

Definitions

  • the present invention relates to a search device using a sensitivity expression word that searches for a proper name using a sensitivity expression word.
  • the first method is to create impression data for each piece of data by measuring the impression given by the individual data in advance using an impression rating experiment using Kansei expression pairs, and input this impression data when searching This is a technique for matching with the sentiment expression word.
  • the second method is a method of matching physical quantities previously assigned to each data based on physical measurement values such as colors and shapes to be searched that satisfy the emotional expression words input at the time of the search.
  • the sensitivity expression words that can be used as search keys are limited to those used in the impression evaluation experiment. For this reason, the user needs to store the emotional expression words that can be used in advance or search for the corresponding emotional expression word from a separately prepared emotional expression word table and use it. For this reason, it has been impossible to perform input to the search device with an expression that matches the sense and sensibility that the user feels with respect to the data to be searched. In addition, when performing matching using impression data, it is necessary to evaluate impressions of all data to be searched in advance, which is extremely difficult when the number of data is large.
  • the physical measurement value extracted based on the emotional expression word does not necessarily sufficiently reflect the similarity to the subjective impression expressed as the emotional expression word. Therefore, the probability of obtaining a search result that satisfies the user's request is low.
  • a Kansei expression word is extracted by a natural language process from a search condition sentence expressed in a natural language as a search condition for image information to be searched.
  • search impression data corresponding to the emotional expression words input by the user is extracted from a subjective evaluation information dictionary in which subjective impressions are given to a plurality of emotional expression words.
  • the impression data integrated by the integration process is set as the search impression data.
  • Impression data for each emotional expression word stored in the subjective assessment information dictionary is a coordinate with each subjective impression element of the impression data as the coordinate axis by specifying the strength of each subjective impression element constituting the impression data. In space, it is determined as a multidimensional coordinate value (vector).
  • image information having impression data that is most similar to the extracted search impression data, that is, having the closest Euclidean distance is output as a search result from the subjective evaluation information dictionary.
  • Patent Literature 1 when a sensitivity expression word extracted from a search condition sentence expressed in a natural language is used as a search key and a degree adverb exists in the search condition sentence, impression data of the corresponding sensitivity expression word Is corrected to meet the degree required by the degree adverb.
  • search results that sufficiently reflect the similarity to subjective impressions, taking into account the phonetic features such as intensities of inflection when the utterance is input by the user's utterance, and the word order of the sensibility expressions There is a problem that cannot be obtained.
  • the present invention has been made to solve the above-described problems, and realizes a search apparatus using a Kansei expression word that can search for a proper name that better reflects a user's search conditions using the word order of Kansei expression words.
  • the purpose is to do.
  • an apparatus for searching for emotional expression words including information indicating a plurality of emotional expression words representing an impression, and a sensitivity vector indicating a degree of relationship with a plurality of typical emotional expression words for each of the emotional expression words.
  • Kansei space database including information to indicate, information indicating a plurality of proper names, and a proper name database including information indicating a sensitivity vector indicating a degree of relationship with a plurality of typical sensitivity expression words for each proper name
  • the character input unit for obtaining character information and the Kansei space database are referred to, the Kansei expression word extracting unit for extracting all Kansei expression words contained in the character information, and the Kansei space database are extracted by referring to the Kansei space database.
  • Kansei vector conversion unit that converts the sentiment expression words into sensitivity vectors, and when the sensitivity vector conversion unit obtains a plurality of sensitivity vectors, the order of the sensitivity expression words included in the character information
  • the proper name is obtained from the proper name database.
  • a unique name search unit for searching and a unique name information output unit for outputting information indicating the unique name searched by the unique name search unit are provided.
  • FIG. 1 It is a figure which shows the function structural example of the search device by the sensitivity expression word which concerns on Embodiment 1 of this invention. It is a figure which shows an example of the sensitivity space database in Embodiment 1 of this invention. It is a figure which shows an example of the proper name database in Embodiment 1 of this invention. It is a figure which shows the hardware structural example of the search device by the sensitivity expression word which concerns on Embodiment 1 of this invention. It is a flowchart which shows the operation example of the search device by the sensitivity expression word which concerns on Embodiment 1 of this invention. It is a figure which shows an example of the language information row
  • FIG. 7A is a diagram showing an example of sensitivity vector information output from the sensitivity vector conversion unit according to Embodiment 1 of the present invention
  • FIG. 7B is a sensitivity output from the sensitivity vector combination unit according to Embodiment 1 of the present invention.
  • FIG. 1 is a diagram showing an example of a functional configuration of a search device 1 using a sensitivity expression word according to Embodiment 1 of the present invention.
  • the retrieval apparatus 1 by a sensitivity expression word includes a sensitivity space database 101, a proper name database 102, a speech input unit 103, a speech recognition unit 104, a prosody information extraction unit 105, a sensitivity expression word extraction unit 106, A vector conversion unit 107, a sensitivity vector combination unit 108, a proper name search unit 109, and a proper name information output unit 110 are provided.
  • the Kansei space database 101 is information (Kansei vector information) indicating a Kansei vector indicating the degree of relationship between information indicating Kansei expression words (Kansei expression word information) and a plurality of representative Kansei expressions for each Kansei expression word.
  • the sensitivity expression word is a character string (language string) that is expressed in a natural language and represents an impression.
  • the representative sensitivity expression word is a typical sensitivity expression word that can express many sensitivity expression words among the sensitivity expression words.
  • the sensitivity vector includes, for example, a value indicating the strength of the relationship between the sensitivity expression word and a plurality of representative sensitivity expression words.
  • FIG. 2 shows an example of the sensitivity space database 101.
  • the Kansei expression word 21 is shown in each row
  • the representative Kansei expression word 22 is shown in each column
  • the corresponding Kansei expression word 21 and the corresponding cell are composed of each row and each column.
  • a value (1 to 5 in FIG. 2) indicating the strength of the relationship with the representative sensitivity expression word 22 is shown.
  • the value of a certain square is 1, it indicates that the relationship between the sensitivity expression word 21 of the row and the dimension (representative sensitivity expression word 22) of the column is weak.
  • the value of a certain square is 5 it indicates that the relationship between the sensitivity expression word 21 in the row and the dimension of the column (representative sensitivity expression word 22) is strong.
  • a value “indicating a relationship between the sensitivity expression word“ calm ”and each of the representative emotion expression words“ fun ”,“ excited ”,“ slow ”, and“ romantic ”. “2”, “1”, “5”, “3” are stored in the corresponding square. This is because the sensibility expression “settled” is strongly related to the representative sensation expression “slow”, and the relationship to the representative sensation expression “romantic” is somewhat strong, and the representative sensation expression “fun” and “excited” "Is a weak relationship.
  • the emotional expression word “settled” in FIG. 2 is associated with the four-dimensional sensitivity vector in which the value indicating the relationship between the emotional expression word and the representative emotional expression word is stored, and stored in the sensitivity space database 101. Is done.
  • the proper name database 102 includes information indicating a plurality of proper names (proprietary name information), and information indicating a sensitivity vector indicating a degree of relationship with a plurality of representative affective expression words for each proper name (sensitivity vector information).
  • the proper name is a character string (language string) or an identification number that represents content (search target) such as a person, a facility, a song, or a moving image.
  • the sensitivity vector includes, for example, a value indicating the strength of the relationship between the proper name and a plurality of representative sensitivity expression words.
  • FIG. 3 shows an example of the proper name database 102.
  • the proper name 31 is shown in each row
  • the representative sentiment expression word 32 is shown in each column
  • the proper unique name 31 and the relevant representative are shown in each cell composed of each row and each column.
  • a value (1 to 5 in FIG. 3) indicating the strength of the relationship with the sensitivity expression word 32 is shown.
  • the value of a certain cell is 1, it indicates that the relationship between the proper name 31 of the row and the dimension of the column (representative emotion expression word 32) is weak.
  • the value of a certain square is 5, it indicates that the relationship between the unique name 31 of the row and the dimension of the column (representative sensitivity expression word 32) is strong.
  • the proper name “ABC Nojima” has a strong relationship with the representative sensibility expression word “romantic”, and the relationship between the representative sensation expression words “fun” and “slow” is somewhat strong, and the representative sensation expression word “excited” "Is a weak relationship.
  • the proper name “ABC Nojima” in FIG. 3 is associated with the four-dimensional sensitivity vector in which the value indicating the relationship between the proper name and the representative sentiment expression word is stored, and held in the proper name database 102.
  • the voice input unit 103 receives voice input and obtains voice information 201. Note that the voice is input from the user to the search device 1 using the emotional expression word.
  • the voice information 201 obtained by the voice input unit 103 is transmitted to the voice recognition unit 104 and the prosody information extraction unit 105.
  • the voice recognition unit 104 performs voice recognition processing on the voice information 201 obtained by the voice input unit 103 and converts the voice information 201 into a language information string (character information) 202 representing the utterance content of the voice information 201.
  • This language information column 202 is character information representing a search condition subjectively expressed by the user.
  • the language information string 202 obtained by the voice recognition unit 104 is transmitted to the prosodic information extraction unit 105 and the emotional expression word extraction unit 106.
  • the prosodic information extraction unit 105 extracts prosody information 203 for the language information string 202 obtained by the speech recognition unit 104 from the speech information 201 obtained by the speech input unit 103.
  • the prosody information 203 includes information (speech feature information) indicating speech features with respect to the language information sequence 202 and information indicating the correspondence between the speech feature information and the language information sequence 202. Examples of the voice characteristics include inflection strength (sound pitch (pitch) and strength (power)), length (tone), speech speed, and at least the inflection strength.
  • the prosodic information 203 extracted by the prosodic information extracting unit 105 is transmitted to the sensitivity vector combining unit 108.
  • the emotional expression word extraction unit 106 refers to the sensitivity space database 101 and extracts all the emotional expression words included in the language information sequence 202 obtained by the speech recognition unit 104. That is, the emotional expression word extraction unit 106 analyzes the language information string 202 by natural language processing, and extracts all phrases that match the emotional expression words included in the emotional space database 101 as sensitivity expression words. Information (sensitivity expression word information 204) indicating the sensitivity expression word extracted by the sensitivity expression word extraction unit 106 is transmitted to the sensitivity vector conversion unit 107.
  • the sentiment vector conversion unit 107 refers to the sentiment space database 101 and converts the sentiment expression word extracted by the sentiment expression word extraction unit 106 into a sentiment vector. That is, the sentiment vector conversion unit 107 converts the sentiment expression word into a sentiment vector associated with the same sentiment expression word included in the sentiment space database 101. In the case where a plurality of sensitivity expression words are extracted by the sensitivity expression word extraction unit 106, the sensitivity vector conversion unit 107 performs conversion into a sensitivity vector for each of the sensitivity expression words. Information (sensitivity vector information 205) indicating the sensitivity vector obtained by the sensitivity vector conversion unit 107 is transmitted to the sensitivity vector combining unit 108.
  • the sentiment vector combining unit 108 when the sentiment vector conversion unit 107 obtains a plurality of sentiment vectors, the inflection strength included in the prosodic information 203 extracted by the prosodic information extraction unit 105 and the sentiment included in the language information sequence 202. Based on the word order of the expression words, a single sensitivity vector is calculated from the plurality of sensitivity vectors. That is, the sensibility vector combining unit 108 calculates the single sensibility vector by combining the sensibility vectors after giving weights based on the level of inflection and word order. Information (sensitivity vector information 206) indicating a single sensitivity vector obtained by the sensitivity vector combining unit 108 is transmitted to the unique name search unit 109.
  • the sensitivity vector combination unit 108 uses the sensitivity vector information 205 from the sensitivity vector conversion unit 107 as the sensitivity vector information 206 as it is and the proper name search unit 109. Send to.
  • the sensitivity vector combining unit 108 obtains the information indicating the word order (word order information) by referring to the prosodic information 203 and the sensitivity vector information 205.
  • the linguistic information string 202 is transmitted to the sensibility vector combination unit 108 in a form associated with the prosodic information 203 (for example, including language information corresponding to each prosodic information 203).
  • the sentiment expression word information 204 is transmitted in the attached form. Therefore, the sensibility vector combination unit 108 can acquire, as the word order, the order in which portions that match the respective sensibility expression words acquired from the sensibility vector information 205 appear in the language information string 202 acquired from the prosody information 203.
  • the sensitivity vector information 205 may be stored according to the appearance order of the sensitivity expression words. In this case, the sensitivity vector combination unit 108 may obtain the word order from the storage order.
  • the proper name search unit 109 searches for a proper name from the proper name database 102 based on the single sentiment vector obtained by the sentiment vector conversion unit 107 or the sentiment vector combination unit 108. That is, the proper name search unit 109 searches the proper name database 102 for a proper name associated with a sensitivity vector similar to the single sensitivity vector. Information (unique name information 207) indicating the unique name searched by the unique name search unit 109 is transmitted to the unique name information output unit 110.
  • the proper name information output unit 110 outputs information indicating the proper name searched by the proper name search unit 109 (proprietary name information 208). Note that the proper name information 208 is output from the search device 1 based on the emotional expression word.
  • the search device 1 based on the emotional expression word composed of the above functions is not limited to Japanese, and may be used in a foreign language such as English.
  • the speech input unit 103 speech recognition unit 104, prosodic information extraction unit 105, Kansei expression word extraction unit 106, Kansei vector conversion unit 107, Kansei vector combination unit 108, proper name search unit 109, and proper name information output unit 110
  • the processor 301 is an arithmetic device that executes a program stored in the program 302.
  • the speech input unit 103, speech recognition unit 104, prosody information extraction unit 105, sensitivity expression word extraction unit 106, sensitivity vector conversion unit 107, sensitivity vector combination unit 108, proper name search unit 109, and proper name information output unit 110 The function is realized by software, firmware, or a combination of software and firmware.
  • Software and firmware are described as programs and stored in the memory 302.
  • the processor 301 reads out and executes the program stored in the memory 302, thereby realizing the functions of the respective units.
  • the emotional expression word search device 1 includes a memory 302 for storing a program that, when executed by the processor 301, for example, causes each step shown in FIG. 5 to be described later to be executed as a result. .
  • These programs include a speech input unit 103, a speech recognition unit 104, a prosody information extraction unit 105, a sensitivity expression word extraction unit 106, a sensitivity vector conversion unit 107, a sensitivity vector combination unit 108, a proper name search unit 109, and a proper name.
  • the computer executes the procedure and method of the information output unit 110.
  • the memory 302 for example, a nonvolatile or volatile semiconductor memory such as a RAM (Random Access Memory), a ROM (Read Only Memory), a flash memory, an EPROM (Erasable Programmable ROM), an EEPROM (Electrically Programmable EPROM), or the like.
  • a magnetic disk a flexible disk, an optical disk, a compact disk, a mini disk, a DVD (Digital Versatile Disc), and the like.
  • the emotional space database 101 and the unique name database 102 are stored in a memory 302 that is a storage device.
  • the voice that is input to the search device 1 by the sensitivity expression word is input by the input interface 303 that is an input device.
  • the unique name information 208 that is output from the search device 1 based on the sensitivity expression word is output by the output interface 304 that is an output device.
  • the processing of the name information output unit 110 may be realized as an electric circuit.
  • step ST501 the voice input unit 103 receives voice input and obtains voice information 201.
  • step ST502 the speech recognition unit 104 converts the speech information 201 obtained by the speech input unit 103 into a language information sequence 202, and the prosodic information extraction unit 105 converts the language information sequence 202 from the speech information 201.
  • the prosodic information 203 is extracted.
  • the prosody information 203 includes at least information indicating the strength of intonation.
  • the voice recognizing unit 104 obtains a language information string “a lively and exciting place” as shown in FIG.
  • the prosodic information extraction unit 105 extracts prosodic information 203 indicating that the voice is spoken at a low volume until “lively”, and the subsequent “exciting place” is spoken at a louder volume than “lively”. .
  • the Kansei expression word extraction unit 106 refers to the Kansei space database 101 and extracts all Kansei expression words included in the language information string 202 obtained by the speech recognition unit 104.
  • the voice recognition unit 104 obtains the language information string “Busy and exciting” shown in FIG.
  • the Kansei space database 101 shown in FIG. 2 includes Kansei expression words “lively” and “exciting”.
  • the emotional expression word extraction unit 106 extracts two phrases “busy” and “exciting” as emotional expression words from the language information string “Busy and exciting” shown in FIG.
  • step ST504 the sentiment vector conversion unit 107 refers to the sentiment space database 101, and converts the sentiment expression word extracted by the sentiment expression word extraction unit 106 into a sentiment vector.
  • step ST505 the sensitivity vector conversion unit 107 determines whether all the sensitivity expression words extracted by the sensitivity expression word extraction unit 106 have been converted into sensitivity vectors. In this step ST505, when the emotion vector conversion unit 107 determines that there is a sensitivity expression word that has not been converted into a sensitivity vector, the sequence returns to step ST504 and repeats the above processing. On the other hand, when the emotion vector conversion unit 107 determines that all the emotion expression words have been converted into the sensitivity vectors, the sequence proceeds to step ST506.
  • the sentiment vector conversion unit 107 extracts the sentiment vector associated with each sentiment expression word from the sentiment space database 101 shown in FIG.
  • the sentiment vector conversion unit 107 converts the sensitivity expression word “lively” into a four-dimensional sensitivity vector (3, 2, 1, 1).
  • “2 3 1 2” is stored in each column of the row corresponding to the sentiment expression word “wakuwaku”.
  • the sensitivity vector conversion unit 107 converts the sensitivity expression word “wakuwaku” into a four-dimensional sensitivity vector (2, 3, 1, 2).
  • the sentiment vector conversion unit 107 transmits the sentiment vector information 205 as shown in FIG. 7A to the sentiment vector combination unit 108.
  • step ST506 the sensitivity vector combining unit 108 determines whether there are a plurality of sensitivity vectors obtained by the sensitivity vector conversion unit 107.
  • the emotion vector combining unit 108 determines that there are not a plurality of sensitivity vectors obtained by the sensitivity vector conversion unit 107, that is, the sensitivity vector information 205 from the sensitivity vector conversion unit 107. Is sent to the proper name search unit 109 as the sensitivity vector information 206 as it is, and the sequence proceeds to step ST510.
  • step ST506 when the emotion vector combining unit 108 determines that there are a plurality of sensitivity vectors obtained by the sensitivity vector conversion unit 107, the sequence proceeds to step ST507.
  • the sensitivity vector combining unit 108 performs sensitivity vector conversion based on the inflection included in the prosody information 203 extracted by the prosody information extraction unit 105 and the word order of the sensitivity expression words included in the language information sequence 202.
  • a weight is assigned to the sensitivity vector obtained by the unit 107.
  • the sensitivity vector combining unit 108 increases the weight when the inflection is strong, and decreases the weight when the intonation is weak. Also, the weight is increased when the word order is early, and the weight is decreased when the word order is later. Details of the processing in step ST507 will be described later.
  • step ST508 the sensitivity vector combining unit 108 determines whether or not weights have been given to all the sensitivity vectors obtained by the sensitivity vector conversion unit 107. In this step ST508, if the emotion vector combining unit 108 determines that there is an emotion vector to which no weight is given, the sequence returns to step ST507 and repeats the above processing. On the other hand, in step ST508, if the perception vector combining unit 108 determines that weights have been assigned to all perception vectors, the sequence proceeds to step ST509.
  • step ST509 the sensibility vector combining unit 108 combines all the sensibility vectors given weights in step ST507, and calculates a single sensibility vector.
  • the sensitivity vector combining unit 108 assigning weights to the sensitivity vectors in the sensitivity vector information 205 illustrated in FIG. 7, the sensitivity vector of the sensitivity expression word “lively” is (2.25, 1.5, 0). .75,0.75), and the sensitivity vector of the sensitivity expression word “wakuwaku” is (1.5, 2.25, 0.75, 1.5).
  • the sensitivity vector combining unit 108 calculates a single sensitivity vector (3.75, 3.75, 1.5, 2.25) by adding the two sensitivity vectors.
  • FIG. 7B shows sensitivity vector information 206 output by the sensitivity vector combination unit 108 in this case.
  • the proper name search unit 109 searches for the proper name from the proper name database 102 based on the single sentiment vector obtained by the sentiment vector conversion unit 107 or the sentiment vector combination unit 108. Specifically, using the single sensitivity vector as a search key, a sensitivity vector that is most similar to the search key is selected from the sensitivity vectors included in the unique name database 102 and is associated with the selected sensitivity vector. Extract unique names.
  • the speech recognition unit 104 obtains the linguistic information string “I want to be excited”
  • the Kansei expression word extraction unit 106 extracts the Kansei expression word “Waku Waku”
  • the Kansei vector conversion unit 107 uses a four-dimensional Kansei vector (2, 3 , 1, 2).
  • the proper name search unit 109 refers to the sensitivity vector of each row included in the proper name database 102 shown in FIG. 3, and the sensitivity vector having the most similar value to the sensitivity vector (2, 3, 1, 2).
  • the unique name associated with is searched. Similar sensitivity vectors are searched by using, for example, the similarity calculated using the cosine distance or the Euclidean distance.
  • the sensitivity vector (2, 3, 1, 2) of the emotional expression word “wakuwaku” has a slightly high “excited” value of “3” and “fun” and “romantic” values of “3”.
  • the value of “slow” is as low as “1”. Therefore, when the proper name search unit 109 searches the proper name database 102 shown in FIG. 3 using this sentiment vector as a search key, the sentiment vector (5, 4, 5) whose “excited” value is as high as “4”. 1, 2), the unique name “HIJ Land” associated with the emotional expression word “Wakuwaku” is extracted.
  • step ST511 the unique name information output unit 110 outputs information indicating the unique name searched by the unique name search unit 109 (proprietary name information 208).
  • the sensitivity vector combining unit 108 receives the prosody information 203 from the prosody information extraction unit 105, and receives the sensitivity vector information 205 from the sensitivity vector conversion unit 107.
  • the sensitivity vector combination unit 108 calculates a time series value of the fundamental frequency F0 of the sound corresponding to the sensitivity expression word from the information indicating the level of intonation included in the prosody information 203, and calculates the average value thereof. calculate.
  • step ST803 the sensitivity vector combining unit 108 determines whether the average value of the fundamental frequency F0 is equal to or greater than a preset threshold value.
  • the sequence proceeds to step ST804.
  • step ST803 if the perception vector combining unit 108 determines that the average value of the fundamental frequency F0 is equal to or greater than the threshold, the sequence proceeds to step ST805.
  • the sensitivity vector combining unit 108 determines the strength of the intonation.
  • the emotion vector combining unit 108 determines that the inflection is strong if the calculated average value of the fundamental frequency F0 is equal to or greater than the threshold value, and determines that the inflection is weak if the average value of the fundamental frequency F0 is less than the threshold value.
  • step ST804 the sensitivity vector combining unit 108 assigns a weight corresponding to the case where the inflection is weak to the sensitivity vector corresponding to the sensitivity expression word. That is, the sensitivity vector combining unit 108 reduces the weight for the sensitivity vector. Also, in step ST805, the sensitivity vector combining unit 108 assigns a weight according to the case where the inflection is strong to the sensitivity vector corresponding to the sensitivity expression word. That is, the sensitivity vector combining unit 108 increases the weight for the sensitivity vector.
  • step ST806 the affective vector combining unit 108 extracts the word order of the affective expression words included in the language information sequence 202.
  • step ST807 the affective vector combination unit 108 determines whether the word order is early.
  • step ST807 when the perceptual vector combination unit 108 determines that the word order is not early, the sequence moves to step ST808.
  • step ST807 when the perceptual vector combination unit 108 determines that the word order is early, the sequence proceeds to step ST809.
  • step ST808 the sensitivity vector combining unit 108 assigns a weight corresponding to the case where the word order is later to the sensitivity vector corresponding to the sensitivity expression word. That is, the sensitivity vector combining unit 108 reduces the weight for the sensitivity vector. Also, in step ST809, the sensitivity vector combining unit 108 assigns a weight corresponding to the sensitivity vector corresponding to the sensitivity expression word when the word order is early. That is, the sensitivity vector combining unit 108 increases the weight for the sensitivity vector.
  • the speech recognition unit 104 obtains the linguistic information string “buzzy and exciting” shown in FIG. 6, and the emotional expression word extraction unit 106 extracts the emotional expression words “lively” and “wakuwaku”, and the sensitivity vector conversion Assume that the unit 107 outputs the sensitivity vector information 205 shown in FIG. Further, it is assumed that the prosodic information extraction unit 105 extracts prosodic information 203 indicating that “lively” has weak inflection and “wakuwaku” has strong intonation. In this case, the sensitivity vector combining unit 108 weights the values (2, 3, 1, 2) of the sensitivity vectors corresponding to the sensitivity expression word “wakuwaku” according to the case where the inflection is strong (for example, 1.5).
  • the sensitivity vector combining unit 108 weights the values (3, 2, 1, 1) of the sensitivity vectors corresponding to the sensitivity expression word “lively” according to the case where the inflection is weak (for example, 0.5). Is assigned (multiplication), and the values of the columns of the sensitivity vector are (1.5, 1, 0.5, 0.5).
  • the emotion vector combining unit 108 compares the average values of the fundamental frequencies F0 corresponding to the respective emotional expression words, and determines the strength of the inflection based on the magnitude. May be.
  • the sensitivity vector combining unit 108 assigns a weight (for example, 0.5) to the value (3, 2, 1, 1) of the sensitivity vector corresponding to the sensitivity expression word “lively” when the inflection is weak. ) Is given (multiplied) to the values (1.5, 1, 0.5, 0.5) given (multiplied), and each weight of the sensitivity vector is given (multiplied). Let the column values be (2.25, 1.5, 0.75, 0.75).
  • the sensitivity vector combining unit 108 weights the values (2, 3, 1, 2) of the sensitivity vectors corresponding to the sensitivity expression word “wakuwaku” according to the case where the inflection is strong (for example, 1.5). Is assigned (multiplied) to the value (3, 4.5, 1.5, 3) obtained by adding (multiplying) to the values (3, 4.5, 1.5, 3). Is (1.5, 2.25, 0.75, 1.5).
  • the sensitivity vector combining unit 108 calculates the weight of the sensitivity vector corresponding to the sensitivity expression word that is in a down tone and has a slow speech speed based on the tone and the speech speed included in the prosody information 203. You may perform the process to make small. It is assumed that the user's certainty is low when the user makes an utterance that expresses hesitation or confusion and is high when the utterance does not express hesitation or confusion.
  • the speech recognition unit 104 obtains the language information string “Romantic ... Hmm, you can relax slowly!”
  • the prosodic information extraction unit 105 indicates that “romantic ... mm” is slow in speaking speed, especially “romantic ...” is in a down tone, and “slow, delicious place!” ⁇ ⁇ ”Is faster, especially“ delicious place! ”
  • Extracted prosodic information 203 indicating that the speech was spoken with strong inflection compared to“ romantic ... ”and“ can do it slowly ”
  • the prosodic information extraction unit 105 indicates that “romantic ... mm” is slow in speaking speed, especially “romantic ...” is in a down tone, and “slow, delicious place!” ⁇ ⁇ ”Is faster, especially“ delicious place! ”
  • Extracted prosodic information 203 indicating that the speech was spoken with strong inflection compared to“ romantic ... ”and“ can do it slowly ”
  • the sensitivity vector combining unit 108 assigns weights to the emotion expression words “romantic”, “slow”, and “delicious” extracted from the language information string 202 based on the prosodic information 203.
  • “romantic ...” is an utterance with a downward tone and a slow speaking speed, so the user is confusing or confused, that is, speaking with a low degree of certainty. Can be determined. Therefore, the sensitivity vector combining unit 108 sets the priority of the sensitivity expression word “romantic” as a word used for the search, that is, reduces the weight of the corresponding sensitivity vector.
  • the sensitivity vector combination unit 108 uses the sensitivity expression word “slow” as a reference for the sensitivity expression word used for the search, and does not assign a weight (the weight is set to “0”). In addition, the sensitivity vector combination unit 108 sets a high priority for the sensitivity expression word “delicious”, that is, increases the weight.
  • the sensitivity vector combining unit 108 performs processing to reduce the weight of the sensitivity vector corresponding to the sensitivity expression word uttered in an upward tone based on the prosodic information 203 or not to add the weight. May be. For example, when an upward utterance such as “romantic?” Is made, it can be determined that the user has expressed an utterance expressing dissatisfaction or disgust. Therefore, the sensitivity vector combination unit 108 does not assign a weight or a weight to the sensitivity expression word “romantic” in which the user expresses dissatisfaction or disgust.
  • the determination of the tone when each sensitivity expression word is uttered in the sensitivity vector combination unit 108 may be determined based on the degree of change in the tone corresponding to the sensitivity expression word, or may be included in the language information string 202. It may be determined by comparing with the tone corresponding to the sensibility expression word.
  • an operation example of weighting to the sensitivity vector based on the tone of the sound and the speech speed by the sensitivity vector combining unit 108 in step ST507 will be described with reference to FIG.
  • the flow shown in FIG. 10 is shown independently from the flow shown in FIG. 8, but the processing shown in FIG. 10 is performed together with the processing shown in FIG.
  • the sensitivity vector combining unit 108 receives the prosody information 203 from the prosody information extraction unit 105, and receives the sensitivity vector information 205 from the sensitivity vector conversion unit 107.
  • step ST1002 the sensitivity vector combining unit 108 calculates a time series value of the fundamental frequency F0 of the sound corresponding to the sensitivity expression word from the prosodic information 203.
  • the sensitivity vector combining unit 108 divides the section of the fundamental frequency F0, and extracts the value of the representative fundamental frequency F0 of each section. At this time, for example, the sensitivity vector combining unit 108 divides the section of the fundamental frequency F0 into two equal parts.
  • the representative value refers to a value such as a maximum value, an average value, or a median value of the fundamental frequency F0 in a section where the fundamental frequency F0 value is extracted.
  • step ST1004 the sensibility vector combining unit 108 calculates a difference in the value of the representative fundamental frequency F0 in each section.
  • the sensitivity vector combining unit 108 subtracts the value of the representative fundamental frequency F0 in the previous section from the value of the representative fundamental frequency F0 in the subsequent section.
  • step ST1005 the emotion vector combining unit 108 determines whether the absolute value of the difference is equal to or greater than a preset threshold value. In this step ST1005, when the emotion vector combining unit 108 determines that the absolute value of the difference is not equal to or greater than the threshold value, the sequence ends. On the other hand, in step ST1005, when the affective vector combination unit 108 determines that the absolute value of the difference is equal to or greater than the threshold, the sequence proceeds to step ST1006.
  • step ST1006 the emotion vector combining unit 108 determines whether the difference is a positive value.
  • step ST1006 when the perceptual vector combining unit 108 determines that the difference value is not positive, the sequence proceeds to step ST1007.
  • step ST1006 when the perceptual vector combining unit 108 determines that the difference value is positive, the sequence proceeds to step ST1009.
  • step ST1007 the affective vector combining unit 108 determines whether or not the utterance time length is greater than or equal to a preset threshold value.
  • the sequence ends.
  • the sequence proceeds to step ST1008.
  • step ST1008 the sensitivity vector combining unit 108 assigns a weight according to the case of the down tone to the sensitivity vector corresponding to the sensitivity expression word. That is, the sensitivity vector combining unit 108 reduces the weight for the sensitivity vector. Also, in step ST1009, the sensitivity vector combining unit 108 assigns a weight according to the upward tone to the sensitivity vector corresponding to the sensitivity expression word. That is, the sensitivity vector combination unit 108 reduces the weight for the sensitivity vector or sets it to zero.
  • the sensitivity vector combination unit 108 determines the tone based on the difference between the values of the representative fundamental frequency F0, both the absolute value of the difference and the threshold value, and the difference value magnitude (positive / negative) are both used. The case of judging was shown. However, the present invention is not limited to this, and the emotion vector combining unit 108 may perform only one of the above determinations.
  • the sensitivity vector combining unit 108 determines the tone using a difference in the value of the representative fundamental frequency F0 in other sensitivity expression words as a threshold value. You may go.
  • the sensibility vector combination unit 108 represents a representative basic extracted from each section of the basic frequency F0 corresponding to the first half “roman” and second half “tick” of the sensibility expression word “romantic”. It is assumed that “romantic” is determined to be in a down tone based on the difference in the value of the frequency F0.
  • the sensibility vector combining unit 108 extracts a difference in the value of the representative fundamental frequency F0 from each section of the fundamental frequency F0 corresponding to the first half “Yu” and the second half “Kuri” of the sensibility expression word “slow”, A “slow” tone is determined by comparison with a difference in the value of the representative fundamental frequency F0 of “romantic”.
  • the difference between the values of the fundamental frequency F0 of “slow” is smaller than that of “romantic”, it can be determined that “slow” is a simple tone.
  • the sensitivity vector combining unit 108 assigns weight values to the sensitivity vectors based on the prosodic information 203 corresponding to the sensitivity expression words when weighting the sensitivity vectors according to the pitch, strength, speech speed, length of the speech, and the like. May be given a preset uniform value, or may be given a value according to the degree of change in pitch, strength, etc. The degree of change such as the level of the sound or the strength may be set by calculating the degree of change from a preset threshold value, or the prosodic information 203 corresponding to other emotional expression words included in the language information string 202. May be calculated and set.
  • information indicating a plurality of sensitivity expression words and information indicating a sensitivity vector indicating a degree of relationship with a plurality of representative sensitivity expression words for each of the sensitivity expression words is extracted from the speech information 201.
  • a prosody information extraction unit 105 a sensitivity expression word extraction unit 106 that refers to the sensitivity space database 101 and extracts all the emotional expression words included in the language information string 202
  • the sensitivity space database 101 is referred to and the sensitivity vector conversion unit 107 that converts the sensitivity expression word extracted by the sensitivity expression word extraction unit 106 into a sensitivity vector
  • the sensitivity vector conversion unit 107 obtains a plurality of sensitivity vectors
  • a sensitivity vector combining unit 108 that calculates a single sensitivity vector from the plurality of sensitivity vectors based on the word order of the sensitivity expression words included in the prosody information 203 and the language information string 202
  • a sensitivity vector conversion unit 107 or a unique name search unit 109 for searching for a unique name from the unique name database 102 based on a single sensitivity vector obtained by the sensitivity vector combining unit 108, and information indicating a unique name searched by the unique name search unit 109
  • a proper name information output unit 110 that outputs the You
  • the proper name database 102 includes information indicating a plurality of proper names and information indicating a sensitivity vector indicating a degree of relationship with a plurality of representative affective expression words for each proper name.
  • the unique name may be a search target, and for example, a song name, a car name, a television program name, a font name, and the like are search targets. Therefore, the unique name database 102 can be easily created, expanded, exchanged, and diverted.
  • the sensitivity vector combining unit 108 increases the weight of the sensitivity vector corresponding to the emotion expression word having a strong inflection, and decreases the weight of the sensitivity vector corresponding to the sensitivity expression word having a weak intonation, Sensitivity vectors can be combined. As a result, a unique name that better reflects the user's intention to speak can be obtained.
  • the sensitivity vector combining unit 108 increases the weight of the sensitivity vector corresponding to the sensitivity expression word with the earlier word order, and decreases the weight of the sensitivity vector corresponding to the sensitivity expression word with the word order later. Sensitivity vectors can be combined. As a result, a unique name that better reflects the user's intention to speak can be obtained.
  • the sensitivity vector combining unit 108 can combine the sensitivity vectors by reducing the weight of the sensitivity vector corresponding to the sensitivity expression word having the falling tone and the slow speaking speed. Thereby, the priority of the sensitivity expression word contained in the search condition expressed by the user's utterance is set, and the proper name reflecting the certainty of the user's utterance can be obtained.
  • the sensitivity vector combining unit 108 can combine the sensitivity vectors without decreasing or giving a weight to the sensitivity vector corresponding to the emotion expression word having an upward tone. Thereby, a search condition is narrowed down from a user's utterance voice, and a proper name reflecting a user's intention can be obtained more accurately.
  • the voice input unit 103 obtains the voice information 201 and the voice recognition unit 104 converts the voice information 201 into the language information string 202.
  • the search device 1 based on emotional expression words may accept input of characters instead of voice.
  • a character input unit that receives a character input and obtains a language information string (character information) 202 is provided.
  • the prosodic information extraction unit 105 is not necessary.
  • the sensitivity vector conversion unit 107 obtains a plurality of sensitivity vectors
  • the sensitivity vector combining unit 108 generates a single from the plurality of sensitivity vectors based on the word order of the sensitivity expression words included in the language information sequence 202.
  • the sensitivity vector is calculated.
  • FIG. 11 is a diagram showing a functional configuration example of the search device 1b based on the sensitivity expression word according to Embodiment 2 of the present invention.
  • the search device 1b using the sensitivity expression word according to the second embodiment shown in FIG. 11 is changed from the search device 1 using the sensitivity expression word according to the first embodiment shown in FIG. 1 to the proper name database 102b.
  • the proper name search unit 109 is changed to the proper name search unit 109b, and a genre extraction unit 111 is added.
  • Other configurations are the same, and the same reference numerals are given and description thereof is omitted.
  • the proper name database 102b includes information indicating a plurality of proper names (proprietary name information), information indicating the genre for each proper name (genre information), and a degree of relationship between a plurality of representative affective expression words for each proper name.
  • Information indicating a sensitivity vector indicating the above.
  • the proper name is a character string (language string) or an identification number that represents content (search target) such as a person, a facility, a song, or a moving image.
  • the genre is a character string representing the classification of the proper name.
  • the sensitivity vector includes, for example, a value indicating the strength of the relationship between the proper name and a plurality of representative sensitivity expression words.
  • FIG. 12 shows an example of the proper name database 102b.
  • the proper name 31 and the genre 33 are shown in each row
  • the representative sentiment expression word 32 is shown in each column
  • the corresponding proper name 31 and the corresponding proper name 31 are shown in each cell.
  • a value (1 to 5 in FIG. 12) indicating the strength of the relationship with the corresponding representative sensibility expression word 32 is shown.
  • the value of a certain cell is 1, it indicates that the relationship between the proper name 31 of the row and the dimension of the column (representative emotion expression word 32) is weak.
  • the value of a certain square is 5, it indicates that the relationship between the proper name 31 of the row and the dimension of the column (representative sensibility expression word 32) is strong.
  • the proper name “ABC Nojima” is associated with the genre “walk”. As described above, the proper name “ABC Nojima” in FIG. 12 includes a genre “walk” for classifying proper names, and a four-dimensional sensitivity vector in which values indicating the relationship between proper names and representative affective expressions are stored. Are stored in the unique name database 102b.
  • the difference between the proper name database 102 in the first embodiment and the proper name database 102b in the second embodiment is that genre information is added to the proper name database 102 in the first embodiment.
  • the genre extraction unit 111 extracts a genre from the language information sequence 202 obtained by the speech recognition unit 104 with reference to the proper name database 102b. That is, the genre extraction unit 111 analyzes the linguistic information string 202 by natural language processing, and extracts a phrase that matches the genre included in the proper name database 102b as a genre. Information indicating the genre extracted by the genre extraction unit 111 (genre information 209) is transmitted to the unique name search unit 109b.
  • the proper name search unit 109b searches for the proper name from the proper name database 102 based on the single sentiment vector obtained by the sentiment vector conversion unit 107 or the sentiment vector combination unit 108 and the genre extracted by the genre extraction unit 111. . In other words, the proper name search unit 109b searches the proper name database 102b for proper names that are associated with sensitivity vectors similar to the single sensitivity vector and are classified into the genre. Information (unique name information 207) indicating the unique name searched by the unique name search unit 109b is transmitted to the unique name information output unit 110.
  • the configuration other than the genre extraction unit 111, the proper name search unit 109b, and the proper name database 102b is the same as that of the first embodiment, and a description thereof will be omitted.
  • the genre extraction unit 111 and the proper name search unit 109b are executed by the processor 301 which is an arithmetic device.
  • the proper name database 102b is stored in the memory 302 which is a storage device.
  • processing of the genre extraction unit 111 and the unique name search unit 109b may be realized as an electric circuit.
  • steps ST1301 to ST1303 are added to the flowchart shown in FIG.
  • the other processes are the same, and the same numbers are assigned and the description thereof is omitted.
  • a case in which the sensitivity space database 101 shown in FIG. 2 and the proper name database 102b shown in FIG. 12 are used is shown.
  • the genre extraction unit 111 refers to the unique name database 102b and extracts a genre from the language information string 202 obtained by the speech recognition unit 104.
  • the speech recognition unit 104 obtains the language information string “moist restaurant” shown in FIG.
  • the genre “restaurant” is included in the proper name database 102b shown in FIG.
  • the genre extraction unit 111 extracts the word “restaurant” from the language information string “moist restaurant” shown in FIG. 14 as a genre.
  • step ST1302 the proper name search unit 109b determines whether the genre is extracted by the genre extraction unit 111, that is, whether the genre information 209 is received. In step ST1302, if the proper name search unit 109b determines that the genre is not extracted by the genre extraction unit 111, the sequence proceeds to step ST510. On the other hand, when the unique name search unit 109b determines that the genre is extracted by the genre extraction unit 111, the sequence proceeds to step ST1303.
  • the unique name search unit 109b creates a unique name from the unique name database 102 based on the single sentiment vector obtained by the sentiment vector conversion unit 107 or the sentiment vector combination unit 108 and the genre extracted by the genre extraction unit 111. Search for a name. Specifically, the single sensitivity vector is used as a search key, the sensitivity vector most similar to the search key is selected from the sensitivity vectors classified into the genre included in the proper name database 102b, and the selected The unique name associated with the sensitivity vector is extracted.
  • the speech recognition unit 104 obtains the language information string “moist restaurant” shown in FIG. 14, the emotion expression word extraction unit 106 extracts the sensitivity expression word “moist”, and the sensitivity vector conversion unit 107 displays the language information string in FIG.
  • the sensibility space database 101 is converted into a four-dimensional sensibility vector (1, 1, 4, 5)
  • the genre extraction unit 111 extracts the genre “restaurant”.
  • the unique name search unit 109b refers to the sensitivity vector classified into the genre included in the proper name database 102b shown in FIG. 12, and is the value most similar to the sensitivity vector (1, 1, 4, 5). The unique name associated with the sensitivity vector having is searched.
  • the sensitivity vector (1, 1, 4, 5) of the sensitivity expression word “moist” has a high “romantic” value of “5” and a “slow” value of “4”.
  • the values of “fun” and “excited” are as low as “1”. Therefore, when the proper name search unit 109b searches the proper name database 102b shown in FIG. 12 using this sensitivity vector as a search key, it is classified into the genre “restaurant” and the value of “romantic” is as high as “5”.
  • the unique name “LMN kitchen” associated with the sentiment vector (2, 1, 4, 5) is extracted as a unique name having a strong relationship with the sentiment expression word “moist”.
  • the proper name database 102b also includes information indicating the genre for each proper name, refers to the proper name database 102b, and extracts a genre from the language information column 202.
  • the unique name search unit 109b includes a single sensitivity vector obtained by the sensitivity vector conversion unit 107 or the sensitivity vector combination unit 108, and the genre extracted by the genre extraction unit 111, from the proper name database 102b. Since the configuration is such that the proper name is searched, when the genre is included in the language information column 202, the proper name in the proper name database 102b to be searched is narrowed down by the genre. You can select a unique name to search from among the narrowed specific names, making it easier for users to speak Ku unique name that has been classified as a genre that reflects it is more possible to accurately search for.
  • the voice input unit 103 and the voice recognition unit 104 are used.
  • the present invention is not limited to this, and a character input unit may be provided instead of the voice input unit 103 and the voice recognition unit 104, as in the first embodiment.
  • the invention of the present application can be freely combined with each embodiment, modified with any component in each embodiment, or omitted with any component in each embodiment. .
  • the device for searching by emotional expression word is capable of searching for a proper name reflecting the user's search condition by using the word order of the sensitivity expression word, and searching by a sensitivity expression word for searching for a proper name by the sensitivity expression word Suitable for use in devices and the like.
  • 1,1b Sensitive expression search device 101 Kansei space database, 102, 102b proper name database, 103 speech input unit, 104 speech recognition unit, 105 prosodic information extraction unit, 106 sensitivity expression word extraction unit, 107 sensitivity vector conversion unit , 108 Kansei vector combination part, 109, 109b proper name search part, 110 proper name information output part, 111 genre extraction part, 201 speech information, 202 language information string, 203 prosodic information, 204 affective expression word information, 205 affective vector information , 206 Kansei vector information, 207 proper name information, 208 proper name information, 209 genre information, 301 processor, 302 memory, 303 input interface, 304 output interface.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Databases & Information Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Machine Translation (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

L'invention concerne un dispositif de recherche basée sur un mot de sentiment qui comporte : une base de données d'espaces de sentiment (101) qui comprend à la fois des informations indiquant une pluralité de mots de sentiment, et des informations indiquant des vecteurs de sentiment qui indiquent chacun le degré de relation entre l'un des mots de sentiment et une pluralité de mots de sentiment représentatifs ; une base de données de noms propres (102) qui comprend à la fois des informations indiquant une pluralité de noms propres, et des informations indiquant des vecteurs de sentiment qui indiquent chacun le degré de relation entre l'un des noms propres et la pluralité de mots de sentiment représentatifs ; une unité d'entrée de caractères qui acquiert une chaîne d'informations de mot (202) ; une unité d'extraction de mot de sentiment (106) qui se réfère à la base de données d'espaces de sentiment (101) et extrait tous les mots de sentiment à partir de la chaîne d'informations de mot (202) ; une unité de conversion de vecteur de sentiment (107) qui se réfère à la base de données d'espaces de sentiment (101) et convertit les mots de sentiment extraits en un ou plusieurs vecteurs de sentiment ; une unité de combinaison de vecteurs de sentiment (108) qui, si l'unité de conversion de vecteur de sentiment (107) a produit une pluralité de vecteurs de sentiment, calcule un seul vecteur de sentiment sur la base de l'ordre des mots de sentiment dans la chaîne d'informations de mot (202) ; une unité de recherche de nom propre (109) qui recherche la base de données de noms propres (102) pour un nom propre sur la base du seul vecteur de sentiment calculé; et une unité de sortie d'informations de nom propre (110) qui transmet des informations indiquant le nom propre trouvé.
PCT/JP2016/088704 2016-12-26 2016-12-26 Dispositif de recherche basée sur un mot de sentiment WO2018122919A1 (fr)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/JP2016/088704 WO2018122919A1 (fr) 2016-12-26 2016-12-26 Dispositif de recherche basée sur un mot de sentiment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/JP2016/088704 WO2018122919A1 (fr) 2016-12-26 2016-12-26 Dispositif de recherche basée sur un mot de sentiment

Publications (1)

Publication Number Publication Date
WO2018122919A1 true WO2018122919A1 (fr) 2018-07-05

Family

ID=62707123

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2016/088704 WO2018122919A1 (fr) 2016-12-26 2016-12-26 Dispositif de recherche basée sur un mot de sentiment

Country Status (1)

Country Link
WO (1) WO2018122919A1 (fr)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111724765A (zh) * 2020-06-30 2020-09-29 上海优扬新媒信息技术有限公司 一种文本转语音的方法、装置及计算机设备

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2001101222A (ja) * 1999-10-01 2001-04-13 Just Syst Corp データ検索装置、及びデータ検索用プログラムが記憶された記憶媒体
JP2006330060A (ja) * 2005-05-23 2006-12-07 Univ Waseda 音声合成装置、音声処理装置、およびプログラム
WO2007148493A1 (fr) * 2006-06-23 2007-12-27 Panasonic Corporation Dispositif de reconnaissance d'émotion
JP2010152862A (ja) * 2008-11-27 2010-07-08 I Business Center:Kk 文章体特定装置およびコンピュータに文章体を特定させるためのプログラム
JP2011048527A (ja) * 2009-08-26 2011-03-10 Nippon Telegr & Teleph Corp <Ntt> 感性情報抽出装置、感性検索装置、その方法およびプログラム
JP2011248020A (ja) * 2010-05-25 2011-12-08 Sony Corp 検索装置、検索方法、及び、プログラム

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2001101222A (ja) * 1999-10-01 2001-04-13 Just Syst Corp データ検索装置、及びデータ検索用プログラムが記憶された記憶媒体
JP2006330060A (ja) * 2005-05-23 2006-12-07 Univ Waseda 音声合成装置、音声処理装置、およびプログラム
WO2007148493A1 (fr) * 2006-06-23 2007-12-27 Panasonic Corporation Dispositif de reconnaissance d'émotion
JP2010152862A (ja) * 2008-11-27 2010-07-08 I Business Center:Kk 文章体特定装置およびコンピュータに文章体を特定させるためのプログラム
JP2011048527A (ja) * 2009-08-26 2011-03-10 Nippon Telegr & Teleph Corp <Ntt> 感性情報抽出装置、感性検索装置、その方法およびプログラム
JP2011248020A (ja) * 2010-05-25 2011-12-08 Sony Corp 検索装置、検索方法、及び、プログラム

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111724765A (zh) * 2020-06-30 2020-09-29 上海优扬新媒信息技术有限公司 一种文本转语音的方法、装置及计算机设备

Similar Documents

Publication Publication Date Title
US20180137109A1 (en) Methodology for automatic multilingual speech recognition
TWI532035B (zh) 語言模型的建立方法、語音辨識方法及電子裝置
JP5480760B2 (ja) 端末装置、音声認識方法および音声認識プログラム
CN103578467B (zh) 声学模型的建立方法、语音辨识方法及其电子装置
TWI539441B (zh) 語音辨識方法及電子裝置
KR100998566B1 (ko) 음성인식을 이용한 언어 번역 방법 및 장치
JP3962763B2 (ja) 対話支援装置
TWI582754B (zh) Language pattern generating means, language pattern generating method and program, voice recognizing apparatus, and voice recognition method and program thereof
US20140350934A1 (en) Systems and Methods for Voice Identification
JP5824829B2 (ja) 音声認識装置、音声認識方法及び音声認識プログラム
WO2016008128A1 (fr) Reconnaissance de la parole à l&#39;aide de la grammaire de mots étrangers
JP2018159788A (ja) 情報処理装置、方法及びプログラム
US12100383B1 (en) Voice customization for synthetic speech generation
JP5073024B2 (ja) 音声対話装置
CN112634865B (zh) 语音合成方法、装置、计算机设备和存储介质
US20230148275A1 (en) Speech synthesis device and speech synthesis method
KR20160104243A (ko) 적어도 하나의 의미론적 유닛의 집합을 음성을 이용하여 개선하기 위한 방법, 장치 및 컴퓨터 판독 가능한 기록 매체
JP7034027B2 (ja) 認識装置、認識方法及び認識プログラム
CN114420159B (zh) 音频评测方法及装置、非瞬时性存储介质
CN112786002B (zh) 一种语音合成方法、装置、设备及存储介质
KR101460447B1 (ko) 외국어 학습을 위한 억양 학습 장치 및 방법
WO2018122919A1 (fr) Dispositif de recherche basée sur un mot de sentiment
CN112397048A (zh) 语音合成的发音稳定性评价方法、装置和系统及存储介质
US12243511B1 (en) Emphasizing portions of synthesized speech
JP6289950B2 (ja) 読み上げ装置、読み上げ方法及びプログラム

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 16925903

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 16925903

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: JP

点击 这是indexloc提供的php浏览器服务，不要输入任何密码和下载