JP2007264284A - Emotion addition device, emotion addition method, and emotion addition program - Google Patents

Abstract

An emotion adding device, an emotion adding method, and an emotion adding program capable of inputting a sentence such as a novel or a story and automatically adding emotion information to the input sentence.
SOLUTION: A sentence input means 26 for inputting a sentence, a morpheme analyzing means 28 for dividing the inputted sentence for each word, and favorable, joyful, dynamic anger, politeness, pleasure, sadness for each word in advance. , Refer to an emotion dictionary that associates emotion information such as static anger, discomfort, grudge, or ordinary, and obtains emotion information for each word of the input sentence, and sentences based on the acquired emotion information Emotion provision means 30 for providing emotion information therein, and emotion-added sentence generation means 32 for generating emotion-added text with emotion information provided by the emotion provision means 30.
[Selection] Figure 1

Description

  The present invention relates to an emotion adding device, an emotion adding method, and an emotion adding program that input an arbitrary sentence, convert the sentence into sound accompanied by an emotion expression, and output the converted voice.

  In recent years, text documents, Internet homepage sentences, blogs, news, and other arbitrary sentences are input and parsed to convert them into sentences containing phonological information. Speech synthesizers that synthesize and output synthesized speech have been developed. In speech synthesis that is generally performed, synthesized speech without emotion expressions such as joy, anger, sadness, discomfort, and resentment is output.

  As an invention for generating synthesized speech with emotional expression, the user can specify a speech mode that expresses emotions such as “Akira”, “Disappointment”, “Angry”, etc., and the speech rate set in advance according to the speech mode There is known an invention in which combinations of levels of voice control parameters such as utterance intensity, voice pitch, and inflection level are collectively set (see, for example, Patent Document 1).

  In the speech synthesizer described in Patent Literature 1, since the synthesized speech representing the emotion is generated using the speech control parameter based on the utterance mode designated by the user, the user can select each speech control parameter. There is no need to set values one by one, and various utterance styles are defined.

  In addition, when the user designates an important part in the text in order to alert the user and read aloud with emotional expression, a prosody that makes the designated important part different from other parts is formed. A speech synthesizer that generates a speech waveform that is read out slowly or with a loud voice is known for the designated important part (see, for example, Patent Document 2).

  In the speech synthesizer described in Patent Document 2, the subject, predicate, number, and proper noun specified by the user are read out slowly or loudly with respect to other text parts. It is possible to synthesize speech that is easy to hear. Furthermore, it is possible to synthesize a voice with a high pitch for bright emotion expressions and a voice with a low pitch for dark emotion expressions.

  Also, in order to generate a more natural synthesized speech, texts with pre-specified emotions such as “anger” and “DJ” and utterance style, and text information that has been subjected to linguistic analysis are added. Acquire, determine local attribute information of each part in the text based on the language analysis result and global attribute information, and generate a synthesized speech signal based on the language analysis result and local attribute information A speech synthesizer is known (see, for example, Patent Document 3).

According to the speech synthesis method described in Patent Document 3, local attribute information of emotion and utterance style in the text is determined from the language analysis result of the input text and global attribute information about the emotion and utterance style. By generating synthesized speech using this information, a more natural synthesized speech can be generated.
Japanese Patent Laid-Open No. 5-100692 Japanese Patent Laid-Open No. 11-231885 JP 2003-302992 A

  In order to output synthesized speech accompanied by emotional expression with the speech synthesizer described in Patent Document 1, it is necessary to designate an utterance mode representing emotions such as “Akira”, “Disappointment”, “Anger” in advance by a user or the like was there. Since sentences such as novels and stories often have different emotions for each sentence, there is a problem that it is necessary to perform speech synthesis by switching the utterance mode for each sentence.

  Also, when synthesizing speech including emotional expressions using the speech synthesizer described in Patent Document 2, there is a problem that the user needs to specify important parts such as a subject, predicate, number, proper noun, etc. there were. In addition, there is a problem that all the same parts of speech in a sentence are synthesized into emphasized speech.

  In addition, when synthesizing speech including emotional expressions using the speech synthesizer described in Patent Document 3, it is necessary to add global attribute information related to emotions and utterance styles in advance in a sentence. was there.

  The present invention has been made to solve the above-mentioned problems, and inputs emotional information automatically for each word or sentence used in the sentence by inputting sentences such as news, novels, and stories. It is an object to provide an emotion addition device, an emotion addition method, and an emotion addition program.

  In order to solve the above-mentioned problem, the emotion adding device according to claim 1 includes a sentence input means for inputting a sentence, a morpheme analyzing means for dividing the inputted sentence for each word, and a favorable, pleasant joy for each word in advance. , Emotional information for each word of the input sentence with reference to an emotional dictionary that associates emotional information such as dynamic anger, politeness, pleasure, sadness, static anger, discomfort, resentment, or ordinary Based on the acquired emotion information, emotion giving means for adding emotion information in the text, and emotion-added text generating means for generating emotion-added text with emotion information given by the emotion giving means It is characterized by providing.

  Further, in the invention according to claim 2, the emotion giving means refers to an emotion dictionary that associates an emotion type and an emotion level as the emotion information for each word, and for each word of the inputted sentence. An emotion type and an emotion level are acquired, and emotion information is added to the sentence based on the emotion type, the emotion level, and the word.

  Further, the invention described in claim 3 is characterized in that the emotion imparting means imparts emotion information for each specific unit in the sentence such as an independent word in the sentence, a phrase, a bracket, or between punctuation marks.

  In the invention according to claim 4, the emotion giving means refers to an emotion dictionary associating an emotion type and an emotion level as the emotion information for each word, and for each word of the inputted sentence. An emotion type and its emotion level are acquired, and emotion information is given in a sentence based on emotion information having the maximum emotion level among the acquired emotion types.

  Further, according to a fifth aspect of the present invention, the emotion imparting means adds an emotion level value for each emotion type based on the emotion information acquired for each word, and among the added emotion types, emotions are added. It is characterized in that the emotion type with the maximum level value and the added emotion level are given in the sentence.

  Further, in the invention according to claim 6, the emotion imparting means acquires emotion emphasis information for each word of the inputted sentence by referring to an emotion dictionary in which words and emotion emphasis information are associated in advance. The acquired emotion emphasis information is added to or multiplied by the emotion level to add emotion information to the sentence.

  The invention described in claim 7 is characterized in that a word stored in association with the emotion emphasis information is a combination of a specific word and a word used in a pair with the specific word.

  The invention according to claim 8 is characterized in that the emotion giving means refers to an emotion dictionary in which words of independent words and emotion information, and words of attached words and emotion information are associated with each other, and the inputted sentence Emotion information is acquired for each word, and if there is different emotion information in the sentence based on the acquired emotion information, the emotion information is given to the sentence with priority on the emotion of the attached word And

  In the invention according to claim 9, the emotion giving means adds the appearance frequency for each emotion type appearing in the sentence based on the emotion information acquired for each word, and Of these, the emotion type having the maximum appearance frequency is acquired as emotion information of the sentence.

  Further, in the invention according to claim 10, in the case where it is not possible to specify one emotion type having the maximum emotion level or maximum frequency from the acquired emotion information, It is characterized in that the emotion type in the order of appearance is added to the sentence.

  Further, in the invention according to claim 11, when it is not possible to specify one emotion type having the maximum emotion level or the maximum frequency from the acquired emotion information, Using the priority order set in advance for the emotion type, the emotion type having a higher priority among the acquired emotion information is given to the sentence.

  Further, in the invention according to claim 12, in the case where it is not possible to specify one emotion type having the maximum emotion level or the maximum frequency from the acquired emotion information, It is characterized in that it is divided into units such as independent words, clauses, and punctuation marks in a sentence and given different emotion types.

  The invention according to claim 13 is characterized in that the sentence input means inputs a plurality of sentences and specifies one emotion type having the maximum emotion level or maximum frequency from the acquired emotion information. If the previous sentence exists before the sentence, the emotion imparting means imparts emotion information of the previous sentence to the sentence.

  Furthermore, the invention according to claim 14 is characterized in that the emotional sentence generating means generates a voice interval or vowel formant for each pronunciation unit such as a phoneme, a syllable, a mora, a morpheme, a word, or an accent phrase. Phonetic information corresponding to the emotion information given by the emotion information giving means with reference to an acoustic dictionary in which the phonetic information such as voice stress, voice rate, voice strength, or intonation and emotion information are associated in advance Is obtained for each pronunciation unit, and voice information is generated.

  The invention described in claim 15 is a sentence input means for inputting a sentence, a morpheme analyzing means for dividing the document into words, an emotion dictionary storing emotion information associated with each word in advance, and an emotion dictionary An emotion addition method using emotion imparting means for providing emotion information in a sentence with reference to and an emotion-added sentence generating means for generating an emotion-added sentence, wherein the sentence is input from the sentence input means; and The morpheme analyzing means divides the input document for each word, the emotion giving means refers to the emotion dictionary to add emotion information in the sentence, and the emotion-added sentence generating means is the emotion Generating a sentence with emotion attached with emotion information given by the granting means.

  The invention according to claim 15 is characterized in that a morpheme analyzing means for dividing a document into words, an emotion giving means for giving emotion information in a sentence, an emotion-added sentence generating means for generating an emotion-added sentence, and the sentence An emotion addition program executed by the information processing means of an emotion addition apparatus comprising an input means, the morpheme analysis means, the emotion giving means, and an information processing means for controlling the emotion-added sentence generating means, The information processing means stores a function of inputting sentences from the sentence input means, a function of dividing the input document into the morpheme analysis means for each word, and emotion information associated with each emotion word in advance in the emotion giving means. A function of referring to a sentiment dictionary and acquiring emotion information for each word, and adding emotion information to a sentence based on the acquired emotion information, Characterized in that to realize the function of generating a sentence with feelings with emotion information provided by the given unit.

  According to the first aspect of the present invention, the input sentence is divided into words, and each word is preliminarily favorable, joyful, dynamic anger, politeness, pleasure, sadness, static anger, discomfort, Referencing an emotional dictionary that associates emotional information such as grudges or normality, so that emotional information is automatically generated for each word in the input text, so text with text is automatically generated. When reading aloud, it is possible to automatically generate speech expressed with emotions according to the content of words in the sentence.

  According to the invention described in claim 2, since the emotion imparting means refers to the emotion dictionary and automatically assigns the emotion type and the emotion level for each word of the sentence, When reading automatically, it is possible to automatically generate speech expressed with various emotions corresponding to the contents of words in the sentence.

  According to the invention described in claim 3, the emotion giving means automatically gives emotion information for each specific unit in the sentence, such as an independent word in the sentence, a phrase, a bracket, and between punctuation marks. Therefore, when a text document or the like is automatically read out, a voice expressed by emotion can be automatically generated for each specific unit in the sentence.

  According to the invention described in claim 4, the emotion imparting means refers to the emotion dictionary, acquires the emotion type and the emotion level for each word of the sentence, and the maximum emotion level among the emotion types. Emotion information is added to the sentence based on the emotion information that has, so when text documents are automatically read out, the voice expressed with strong emotion corresponding to the contents of the sentence is automatically generated. Can do.

  According to the invention of claim 5, the emotion imparting means adds the emotion level value for each emotion type, and the emotion type of the maximum emotion level among the added emotion types is included in the text. Since the sentence is automatically assigned to the sentence, the sentence can be read out by voice using the emotion most strongly expressed in the sentence.

  According to the invention described in claim 6, the emotion imparting means obtains emotion emphasis information for each word of the sentence with reference to the emotion dictionary, and adds or multiplies the emotion emphasis information to the emotion level. Since emotion information is given in the text, if there is a word that emphasizes the emotion in the text, the text can be read out with a voice expressed by emphasizing the emotion according to the content of the text.

  Further, according to the invention described in claim 7, as a word stored in association with emotion emphasis information, a combination of a specific word and a word used in a pair with the specific word is used. Emotion emphasis information corresponding to a combination of an adverb and a word corresponding to the adverb, a combination of a compound verb and a word corresponding to the compound verb, and the like can be given to the sentence.

  With this, for example, “What” for a sentence with a combination such as “What do you think?” Or “What?” It's unclear whether it's just an angry emotion or a gently speaking emotion, and it's unclear whether it's just "What?" Can be added.

  According to the invention described in claim 8, the emotion giving means refers to the emotion dictionary in which the words of the independent words and the emotion information and the words of the attached words and the emotion information are associated with each other, and different emotion information in the sentence If there is, the emotion information is added to the sentence giving priority to the emotion of the attached word, so that it is possible to generate a sentence with emotion more accurately adding the emotion information. By automatically reading out the sentence with emotion, it is possible to automatically generate a voice that expresses the contents of the sentence more accurately.

  According to the invention of claim 9, since the emotion imparting means adds the appearance frequencies of the respective emotion types and acquires the emotion type with the highest appearance frequency as the emotion information of the text. It is possible to automatically generate speech expressed by emotions that are used most frequently.

  According to the invention of claim 10, when it is not possible to specify one emotion type having the maximum emotion level or maximum frequency from the acquired emotion information, the emotion imparting means appears. Since the last emotion type is added to the sentence, it is not possible to identify the single emotion that has the maximum emotion level or maximum frequency even though there are multiple types of emotion information in the sentence. Even in this case, it is possible to generate a sentence expressed by emotion.

  According to the invention of claim 11, when it is not possible to specify one emotion type having the maximum emotion level or maximum frequency from the acquired emotion information, the emotion imparting means acquires Since the emotion type with high priority among the plurality of sentiment information is given in the sentence, for example, favorable> joy> dynamic anger> politeness> pleasure> sadness> static anger> discomfort> It is possible to acquire emotion information with a high priority by using a predetermined priority order such as itching> normal.

  As a result, even if there are multiple types of emotion information in the sentence, even if it is not possible to identify one emotion that has the maximum emotion level or maximum frequency, Can be generated.

  According to the twelfth aspect of the present invention, when it is not possible to specify one emotion type having the maximum emotion level or maximum frequency from the acquired emotion information, the emotion imparting means , Separated by units such as independent words, clauses, punctuation marks, etc., and added different emotion types, so even if there are multiple types of emotion information in the sentence, the maximum emotion Even if it is not possible to specify a single emotion that has a level or maximum frequency, it is possible to generate a sentence expressed in emotion for each unit such as a word, a phrase, or between punctuation marks.

  Further, according to the invention of claim 13, it is a case where one emotion type having the maximum emotion level or maximum frequency cannot be specified from the acquired emotion information, and before the sentence. When the previous sentence exists, the emotion imparting means adds the emotion information of the previous sentence in the sentence, so even though there are multiple types of emotion information in the sentence, Even if it is not possible to identify one emotion that has the maximum emotion level or maximum frequency, it is possible to generate a sentence expressed by an emotion linked to the emotion of the previous sentence.

  According to the invention described in claim 14, the emotion-added sentence generating means generates a sound pitch when generating sound for each pronunciation unit such as a phoneme, a syllable, a mora, a morpheme, a word, or an accent phrase. Since voice information is generated for each pronunciation unit by referring to an acoustic dictionary in which emotion information is pre-associated with phonological information such as vowel formant, voice stress, utterance speed, utterance intensity, or intonation, etc. It is possible to automatically generate a voice that expresses emotion for each word.

  According to the invention described in claim 15, the input sentence is divided for each word, and emotion information is obtained for each word of the input sentence by referring to an emotion dictionary in which emotion information is associated with each word in advance. Since the sentence with emotion is automatically generated, when the text sentence or the like is automatically read out, it is possible to automatically generate the voice expressed with the emotion according to the content of the word in the sentence.

  According to the invention described in claim 16, the inputted sentence is divided into words, and emotion information is obtained for each word of the inputted sentence by referring to an emotion dictionary in which emotion information is associated with each word in advance. Since the sentence with emotion is automatically generated, when the text sentence or the like is automatically read out, it is possible to automatically generate the voice expressed with the emotion according to the content of the word in the sentence.

  The best mode for carrying out the present invention will be described below with reference to the drawings.

  First, with reference to FIG. 1, the structure of the signal processing system block of the emotion addition apparatus 900 of this invention and the function of each block are demonstrated.

  The emotion adding device 900 inputs novels, stories, and other sentences via, for example, the Internet 10, performs morphological analysis on the input sentences, and divides them into units of dictionary headings of the emotion dictionary 22, and for each unit. A device such as a computer device or a terminal device that generates sentences that relate emotion information such as favor, joy, dynamic anger, politeness, pleasure, sadness, static anger, discomfort, resentment, or ordinary is there. Furthermore, the emotion adding device 900 has a function of performing voice synthesis using emotion information associated with a sentence and outputting a synthesized voice in which emotion is expressed.

  The emotion adding device 900 transmits / receives information to / from other communication devices as well as an antenna 960 for transmitting / receiving radio waves to / from a carrier such as a wireless LAN or a mobile phone as a communication means. , News or text posted on a homepage, and other wireless communication means 961 that can acquire various other texts, and information is transmitted to and received from other communication devices via a wired communication network such as the Internet 10. In addition, it includes wired transmission / reception means 962 capable of acquiring novels, stories, news, sentences posted on homepages, and other various sentences from other communication devices.

  The emotion adding device 900 includes an input unit 970 including a keyboard, a mouse, a trackball, a tablet, and the like, and characters and various instruction information input from the user by the input unit 970 via a bus 999. An input interface 971 for transmitting information to the information processing means 980, a display means 972 for displaying characters and images to the user, and display information from the information processing means 980 via the bus 999 to obtain video A display interface 973 that converts the signal into a signal and outputs the signal to the display means.

  The emotion adding device 900 stores various texts, emotion dictionaries, acoustic dictionaries, and various processing programs, and can also be used as a work area for the information processing means 980, such as a storage means 984 such as a hard disk or a semiconductor memory, The RAM 981 that can temporarily store various processing programs such as and constants, and can be used as a work area of the information processing means 980, and the BIOS (basic input output system) executed by the information processing means 980, constants, etc. A ROM 983 for storing various information and a timer 990 for measuring time are provided. Note that various texts, emotion dictionaries, acoustic dictionaries, and various processing programs may be stored in the RAM 981 or the ROM 983.

  The emotion adding device 900 obtains voice information synthesized when the text is read out from the information processing means 980, converts the voice information into a voice signal, amplifies and outputs it, and outputs the voice signal as air vibration. And a sound output means 965 such as a speaker for converting the sound into sound.

  The emotion adding device 900 also includes wireless or wired communication processing, processing of input information from the input means 970, display processing for displaying various information on the display means 972, information for the storage means 984, RAM 981, ROM 983, and the like. Writing or reading processing, processing for inputting a sentence from the storage means 984 or other communication device, processing for dividing the sentence for each dictionary heading unit of the emotion dictionary 22, processing for generating a sentence in which emotion information is associated with a word, sentence Processing to acquire emotion information every time, processing to generate text with emotion by associating text and emotion information, processing to generate text with phoneme by associating phonemic information for each pronunciation unit, and generation of speech signal for each pronunciation unit Information processing means 980 for executing the processing to be performed or other various processing based on the processing program. A CPU or MPU can be used as the information processing means 980, and an element such as PAL can be used for simple processing.

  The information processing means 980 of the emotion adding device 900 and its peripheral circuits are connected by a bus 999 and can transmit information at high speed. The information processing unit 980 can control peripheral circuits based on an instruction of a processing program that operates in the information processing unit 980. The processing program executed by the information processing unit 980 can be provided by a recording medium or communication. In addition, each peripheral circuit can be configured by an ASIC or the like.

  Hereinafter, each function that the information processing unit 980 can realize based on a program will be described.

[About text input means 26]
The text input means 26 is based on the text stored in the storage means 984 such as a hard disk (including the function of the text storage means 20), the wireless transmission / reception means 961 or the wired transmission / reception means 962 based on the designation by the user. A function of inputting a sentence from the sentence storage means 20 of another communication device is provided. The function of the text input unit 26 may be realized using a program executed by the information processing unit 980 or may be realized using a dedicated processing circuit.

[About the morphological analysis means 28]
The morpheme analyzing means 28 has a function of separating input sentences such as novels and stories. The morpheme analysis unit 28 divides sentences in units of dictionary headings (see FIGS. 2A, 2B, and 3) of the emotion dictionary 22, and hereinafter, what is divided by the morpheme analysis unit 28 is referred to as a word. This word corresponds to the word in the claim.

  The function of the morphological analysis unit 28 may be realized using a program executed by the information processing unit 980, or may be realized using a dedicated processing circuit.

[Emotion giving means 30]
Emotion giving means 30 associates emotion information such as favor, joy, dynamic anger, politeness, pleasure, sadness, static anger, discomfort, resentment, or ordinary for each word (dictionary headline) in advance. The emotion dictionary 22 is referred to, emotion information is acquired for each word divided by the morphological analysis means 28, and emotion information is provided in the sentence based on the acquired emotion information.

  Further, the emotion imparting means 30 refers to the emotion dictionary 22 in which the emotion type and the emotion level are associated as emotion information for each word delimited by the morpheme analyzing means 28, and for each word delimited by the morpheme analyzing means 28. Is provided with a function of acquiring the emotion type and the emotion level, and adding emotion information to the sentence based on the emotion type, the emotion level, and the words delimited by the morphological analysis means 28.

  The emotion imparting means 30 has a function of imparting emotion information for each specific unit in the sentence, such as an independent word in the sentence, a phrase, a bracket, or between punctuation marks.

  Also, the emotion imparting means 30 refers to the emotion dictionary 22 associating the emotion type and the emotion level as emotion information, and for each word (dictionary heading) delimited by the morphological analysis means 28, the emotion type and the emotion level. And the function of adding emotion information in the sentence based on emotion information having the maximum emotion level among the acquired emotion types.

  Also, the emotion imparting means 30 uses the emotion information acquired for each word delimited by the morpheme analyzing means 28, and among the emotion types obtained by adding the emotion level numerical value for each emotion type, the emotion level numerical value. Has the function to give the emotion type that gives the maximum and the added emotion level to the sentence.

  In addition, the emotion imparting means 30 refers to the emotion dictionary 22 in which the word (dictionary headline) and the emotion emphasis information are associated in advance, and acquires emotion emphasis information for each word (dictionary headline) delimited by the morphological analysis means 28. In addition, it has a function of adding or multiplying the acquired emotion emphasis information to the emotion level to add emotion information to the sentence.

  Further, the emotion imparting means 30 refers to the emotion dictionary 22 that associates the words of independent words (dictionary headings) and emotion information, and the words of adjunct words (dictionary headings) and emotion information, and the morpheme analyzing means 28 Emotion information is acquired for each delimited word (dictionary heading), and if there is different emotion information in the sentence based on the acquired emotion information, the emotion information in the sentence is given priority on the emotion of the attached word It has a function to give.

  In addition, the emotion imparting means 30 has a function of adding the appearance frequency for each emotion type appearing in the sentence and acquiring the emotion type having the maximum appearance frequency among the added emotion types as the emotion information of the sentence. I have.

  In addition, when the emotion imparting means 30 cannot identify one emotion type having the maximum emotion level or maximum frequency from the acquired emotion information, the emotion imparting means 30 selects the emotion type in the order of appearance. It has a function to be given in the text.

  In addition, when the emotion imparting unit 30 cannot identify one emotion type having the maximum emotion level or the maximum frequency from the acquired emotion information, It has a function to give emotion types with high priority to sentences.

  In addition, when the emotion imparting means 30 cannot identify the maximum emotion level or one emotion type having the maximum frequency from the acquired emotion information, the independent words, phrases, It has a function that separates punctuation marks and gives different emotion types.

  Also, the emotion imparting means 30 is a case in which one emotion type having the maximum emotion level or the maximum frequency cannot be specified from the acquired emotion information, and the preceding sentence before the sentence. Is present, the emotion information of the previous sentence is provided in the sentence.

  Each function of the emotion imparting means 30 may be realized using a program executed by the information processing means 980, or may be realized using a dedicated processing circuit.

[About the emotion-added sentence generating means 32]
The emotion-added text generating unit 32 has a function of generating the emotion-added text with emotion information provided by the emotion applying unit 30.

  In addition, the emotion-added sentence generating unit 32 generates a voice pitch or vowel formant, voice stress, and voice rate when generating a voice for each pronunciation unit such as a phoneme, syllable, mora, morpheme, word, or accent phrase. The phonetic information corresponding to the emotion information given by the emotion information giving means is acquired for each pronunciation unit by referring to the acoustic dictionary 24 in which the phonetic information such as utterance intensity or intonation and emotion information are associated in advance. It has a function to generate voice information.

  Note that the acoustic dictionary 24 has a voice pitch or vowel formant, voice stress, voice speed, voice strength, when generating voice for each pronunciation unit such as phonemes, syllables, mora, morphemes, words, or accent phrases. Or a dictionary in which phoneme information such as intonation and emotion information are associated in advance.

  The function of the emotion-added sentence generating unit 32 may be realized by using a program executed by the information processing unit 980 or may be realized by using a dedicated processing circuit.

  Next, the configuration of the emotion dictionary 22 in which the emotion imparting means 30 refers to acquire emotion information will be described.

  FIG. 2A and FIG. 2B are diagrams showing the configuration of the independent word emotion dictionary 22 in which the independent word and the emotion information are associated, and FIG. It is a graph which shows the structure of the emotion dictionary 22 of the attached word which linked | related. FIG. 4 is a chart showing a legend of symbols used as emotion information.

  In the example shown in FIG. 2 and FIG. 3, the embodiment is described in which the independent word emotion dictionary and the adjunct emotion dictionary are separated, but priority is given to the independent word in the sentence and the adjunct word in the sentence. In the case where the ranking is not determined, one emotion dictionary may be used in which the emotion dictionary of independent words and the emotion dictionary of attached words are merged without being divided.

  As shown in FIGS. 2 (a), 2 (b), and 3, the emotion dictionary 22 includes a dictionary heading for searching emotion information and the like of each word in which sentences are separated, and a word corresponding to the dictionary heading. Emotion information of words composed of emotion type and emotion level of words, emotion emphasis information used when emphasizing emotions in sentences, reading of words used when reading out words automatically, and reading out words automatically Accent information used at the time and main parts of speech are stored in association with each other.

  For example, in the emotion dictionary 22 shown in FIGS. 2A and 2B, a word having a dictionary heading “that bastard” is registered. Since this word can be defined as expressing "dynamic anger", the symbol "B" indicating the emotion expressing "dynamic anger" is stored as an emotion type in association with "dictionary headline". ing. Also, since the word “that bastard” is recognized as an expression with a high level of emotion among the emotional expressions of “dynamic anger”, “2” is given as the emotion level, and emotion information “B2” is assigned to the “dictionary”. It is stored in association with “Heading”.

  In addition, in the case of the word having the dictionary heading “Ichiban”, it is not recognized that the word contains a special emotion, and therefore no particular emotion type is associated. However, since the adverb “first” is recognized as an expression that emphasizes another word, emotion emphasis information “+2” for increasing the emotion level of the corresponding other word by two is added to the “dictionary heading”. It is stored in association with. In the example of the word “most”, an embodiment is used in which “2” is added to the emotion level of another corresponding word, but the emotion expression is emphasized using a value that is multiplied by the emotion level of the other word. You may make it do.

  In addition, in the case of a word having a dictionary heading of “What”, the emotion emphasis information can be changed in accordance with the corresponding word. For example, when “what, where, how many” words exist within the last three phrases as the word corresponding to “what”, “+2” is adopted as emotion emphasis information, and “+1” when there is no word. Is stored in association with the “dictionary heading”.

  Further, in the attached word emotion dictionary 22 shown in FIG. 3, a symbol of a dictionary heading “!” Is registered. Since this symbol is not recognized as expressing a specific emotion, it associates emotion emphasis information of “+1” with “dictionary headline” to raise the emotion level of the whole sentence by one without associating the emotion type. I remember. The symbol “!” Uses an embodiment in which “1” is added to the emotion level of the sentence, but the emotion expression may be emphasized using a value that is multiplied by the emotion level of the sentence.

  Also, if the sentence ends with the word “i”, the word can be defined as an adjective ending that expresses “joy”, so a symbol indicating that it is an emotion that expresses “joy” “A” is stored as an emotion type in association with “dictionary headline”. In addition, when “-i” is used at the end of the sentence, “1” is given as the emotion level because the emotion level of “joy” is recognized as an expression with a low emotion level. Is stored in association with the “dictionary headline”.

  Also, if the word ends with the word “I”, it is not recognized that the adjective ending word contains special emotions, so “F0” representing “ordinary” emotion is expressed as emotion information “ It is stored in association with the dictionary heading.

  Also, when the sentence ending is a compound verb ending “Teruda”, the emphasis information can be changed according to the corresponding word. For example, if “what, where, how many” words exist within the previous three paragraphs as the word corresponding to “terudan”, “B2” representing strong “dynamic anger” is expressed as emotion information. If it does not exist, an instruction to adopt the emotion information “A1” indicating favor or pleasure is stored in association with the “dictionary headline”.

  FIG. 4 is a chart of an embodiment showing the types of emotions and the types of emotion levels that the emotion information symbols used in FIGS. 2 (a), 2 (b), and 3 mean.

  In the example shown in FIG. 4, the emotion type “A” is determined to represent a word expressing “favorite” or “joy” such as “smile” or “Jahn”, and the emotion type “B” is It is determined that it represents a word expressing "dynamic anger" such as "roar" or "killing", and the emotion type "C" is "polite" or "pleasant" such as "eating" or "we are" It is determined to represent a word expressing.

  Also, the emotion type “D” is determined to represent a word expressing “sadness” such as “succulent”, and the emotion type “E” is set to “slow”, “poor”, or “muddy”. Such as “Ru” and “I” that use the emotion type “F” at the end of a word that expresses “static anger”, “discomfort”, and “itch” It is defined as a word that expresses emotion.

  Also, as shown in FIG. 4, an emotion level is added to each of the emotion types “A” to “F”. In the example shown in FIG. 2A, FIG. 2B, and FIG. 3, the emotion level “0” indicates that the emotion level expressed by the word is the weakest level, and the emotion level “5” The emotion level expressed by the word is the strongest level. When the emotion expressed by the word is “normal”, it is recognized that there is no difference in the emotion level of “normal”, so only “0” is used as the emotion level.

  Next, processing executed by the emotion adding device 900 will be described using a flowchart.

  FIG. 5 is a flowchart of a process of inputting a novel, a story, news or other text, associating emotion information using a word of the input text, and automatically generating sound expressed based on the emotion information. It is.

  In step S104 “Specify text to be read” (hereinafter abbreviated as S104), the user operates the input means 970 of the emotion adding device 900 and the like to specify information for automatically specifying the text to be read. input. This input may be input by the user directly operating the input unit 970 of the emotion adding device 900, or other communication via the communication network such as the Internet 10 and the wired transmission / reception unit 962 or the wireless transmission / reception unit 961. You may make it input designation | designated of the reading text from an apparatus.

  When a sentence to be read out automatically is designated by a user or the like, the designation is acquired by the information processing means 980 via the bus 999. When the designation of the reading text is input, the information processing unit 980 (including the function of the text input unit 26) inputs the specified series of texts from the text storage unit 20 in S204 “Input specified text”. Process. The text storage unit 20 may be provided inside the storage unit 984 such as a hard disk, or may be provided in another communication device or server connected via the Internet 10 or the like. When the input of the designated series of sentences is completed in S204, the process executed by the information processing unit 980 proceeds to the process of S206 “single sentence analysis”.

  In S <b> 206, the information processing unit 980 performs processing for analyzing the input sentence and dividing it into sentences. The sentence delimiter is not limited to each sentence, but may be delimited for each paragraph, each sentence, each bracket, and each punctuation mark. When the one sentence analysis process is completed, the process executed by the information processing means 980 proceeds to the next step S208 “separate sentences into words”.

  In S208, the information processing unit 980 performs processing such as division on the sentence, and performs processing for dividing the sentence into words. When the process of dividing the sentence into words is completed, the process executed by the information processing unit 980 proceeds to the next S210 “emotion provision” process.

  In S210, the information processing means 980 refers to the emotion dictionary 22 using each divided word, and performs processing for acquiring information related to words such as emotion information, reading, accent, or part of speech. As shown in FIG. 2A, FIG. 2B, and FIG. 3, the emotion dictionary 22 stores in advance, for each word, favor, joy, dynamic anger, politeness, pleasure, sadness, static Emotion information such as anger, discomfort, grudge, or ordinary is stored in association with each other. For example, when acquiring emotion information for the word “beauty”, the emotion type of “C” (see FIG. 4) associated with the word “beauty”, the emotion level of “1”, and the part of speech A process of searching and acquiring from the emotion dictionary 22 is performed (see FIG. 2A). And the process which produces | generates the sentence which provided the emotion was linked | related with the emotion information acquired about all the words.

  For example, as a sentence of Example 1, using “Mirror mirror, who is the most beautiful in the world?” (Quoted from Grimm Fairy Tale: Snow White) and generating a sentence with emotions for each word, Mirror (noun) / yo (particle) | Mirror (noun) /, (symbol) | This world (noun) / de (particle) | Ichiban (adverb [+2]) | Beauty (adjective [C1]) / F0] / noha (particle) | daare (noun) /? (Symbol) / "is generated. When the generation process of the sentence to which the emotion is given ends, the process executed by the information processing unit 980 proceeds to the next step S212 “Acquire emotion information for each sentence”.

  In S212, the information processing unit 980 (including the function of the emotion-added sentence generating unit 32) performs processing for generating a sentence with emotion information based on each emotion information of the sentence to which the emotion is given.

For example, in the case of the above sentence, “Mirror and mirror, who is the most beautiful in the world?”, “+2” emotion emphasis information is associated with the word “Most” and “Beauty” The emotion type [C1] and the emotion level are associated with the word. For the word “I”, the emotion type and emotion level of [F0] are associated with each other. As a representative process of determining emotion information used when automatically reading this sentence,
(1) Processing for obtaining the emotion type having the maximum emotion level and the emotion level among the emotion information in the sentence as emotion information of the sentence (2) Adding each emotion level in the sentence for each emotion information (3) A process of acquiring the emotion type most frequently used in the sentence as the emotion information of the sentence sentence. (4) Process of acquiring emotion information of sentence using emotion information of attached word in sentence (5) Process of inheriting emotion information of previous sentence (6) Out of each emotion information in sentence, appearance order is backward To obtain the emotion type and the emotion level as the emotion information of the sentence (7) Of the emotion information in the sentence, the priority set in advance is used, and the emotion type with the higher priority and the emotion level are sentenced Emotions Treatment processing for acquiring may be mentioned as.

  Also, if it is not possible to identify one emotion type with the maximum emotion level or maximum frequency from each emotion information in the sentence, it is divided in units such as words, phrases, punctuation marks in the sentence, It is also possible to perform a process of assigning individual emotion types for each unit.

  The description of the processes (1) to (7) and the details of the process of S212 will be described separately using the flowchart of FIG. 6, and here, the embodiment using the process (1) is used. I will explain it.

  Of the emotion types of the sentence in the first embodiment, the emotion information having the maximum emotion level is [C1] for the word “beauty”. Since the sentence of Example 1 has [+2] emotion emphasis information associated with the word “Ichiban” in addition, [C1] of “Beauty” and [+2] of “Ichiban” It is possible to acquire emotion information of [C3]. When the emotion information of the text is acquired, the information processing unit 980 (including the function of the emotion-added text generation unit 32) performs a process of generating an emotion-added text that associates the emotion information with the text.

  For example, in the case of the sentence of Example 1 described above, the sentence with emotion generated here adds [C3] to the head of the sentence “[C3]: mirror (noun) / yo (particle) | mirror (noun) /, (Symbol) | this world (noun) / de (particle) | first (adverb [+2]) | beautiful (adjective [C1]) / i [F0] / noha (particle) | daare (noun) /? (Symbol) / ".

  When the generation of the sentence with emotion is finished, the processing executed by the information processing unit 980 proceeds to S214 “Generation of sentence with phoneme”.

  In the next S214, the information processing means 980 (including the function of the emotion-added sentence generating means 32) uses the word reading of the sentence with emotion, the accent, the part of speech, and the emotion information, and refers to the acoustic dictionary 24 to pronounce. A phonological sentence with associated phoneme information is generated for each unit. For the acoustic dictionary 24, reading is specified for each pronunciation unit such as phonemes, syllables, mora, morphemes, words, accent phrases, etc., and the pitch or vowel formant of the speech when generating the speech, the strength of the speech, the utterance The phoneme information such as speed, utterance intensity, or intonation is acquired, and a process for generating a phoneme-added sentence for uttering speech is performed. When the generation of the phoneme-added sentence is completed, the process executed by the information processing unit 980 proceeds to the next process of S216 “generation of voice information”.

  In S216, the information processing unit 980 generates speech information for generating a speech waveform based on the phoneme-added sentence generated in S214. Then, the audio information is output to the output audio processing unit 966, and a process of producing a voice with emotion expression from the audio output unit such as a speaker is performed.

  In the example of the sentence of the first embodiment, a voice with emotion added is uttered by reading “Kagamyo | Kagami | Konoyode / Ichiban / Utsukushinowa / Dare”. The voice including emotional expression is received by the user in S106 “voice listening”.

  The voice information is generated based on phonetic information such as voice pitch or vowel formant, voice stress, utterance speed, utterance intensity, or inflection, etc. Generate adjusted audio information. The generation of the voice information or the voice signal may be all synthesized electronically, or may be generated by synthesizing a voice produced by a voice actor based on emotional expression and disassembled for each pronunciation unit.

  In addition, when generating speech information, the phonetic information for words that have emotion information associated with them is read aloud slowly or according to the emotion type, emotion level, etc. To obtain voice information.

  When the voice information generation process ends, the process executed by the information processing means 980 proceeds to the next S220 “last sentence?”.

  In S220, the information processing means determines whether or not the sentence for which the voice information is generated in S216 is the last sentence in the series of sentences input in S204. When the sentence still remains (S220; No), the process executed by the information processing unit 980 returns to S206, and when the sentence for which the voice information is generated in S216 is the last sentence (S220; Yes). ), The process executed by the information processing means 980 proceeds to “End” in S222, and the emotion addition process ends.

  Next, detailed processing of S212 “acquire emotion information for each sentence” will be described with reference to the flowchart of FIG.

  When the process proceeds to S212 in FIG. 5, the process executed by the information processing unit 980 (including the function of the emotion imparting unit 30 or the emotion-added sentence generating unit 32) is S302 “acquires emotion information of the sentence from the words in the sentence. The process proceeds to.

  In S302, the information processing means 980 obtains the emotion type having the maximum emotion level and the emotion level as emotion information of the sentence among the emotion information associated with each word in the sentence (1). To generate a sentence with emotion.

  When the process (1) is used, a process of acquiring the emotion type having the maximum emotion level and the emotion level among the emotion information of each word in the sentence as the emotion information of the sentence is performed. In the case of the sentence of the first embodiment, there is emotion information of [C1] for the word “beauty” and [F0] for the word ending with “I”, but the emotion level “ The emotion information of [C1] for the word “beauty” is acquired. Further, when the emotion emphasis information is adopted, the [+2] emotion emphasis information associated with the word “most” is adopted, and “C1” of “beauty” and “ +2] is used to obtain emotion information of [C3].

  Further, instead of the process (1), the processes (2) to (7) may be performed to generate a sentence with emotion.

  When the process (2) is used, the emotion level of each word in the sentence is added for each emotion type, and the emotion level having the maximum emotion level and the added emotion level is acquired as the emotion information of the sentence. Perform the process. In the case of the sentence of the first embodiment, there is only emotion information of [C1] for the word “beauty” and [F0] for the word ending with “I”. The emotion information of [C1] for the word “shi” is acquired. Further, when the emotion emphasis information is adopted, the [+2] emotion emphasis information associated with the word “most” is adopted, and “C1” of “beauty” and “ +2] is used to obtain emotion information of [C3]. In the case of the text of Example 1, the same text with emotion is obtained as in the case of using the process (1).

  When the process (3) is used, the emotion type associated with each word in the sentence is added for each emotion type, and the most frequently used among the added emotion types. A process of acquiring the emotion type having the highest frequency as the emotion information of the sentence is performed. In the case of the sentence of Example 1, there is one emotion information of [C1] for the word “beauty” and [F0] for the word ending with “I”, so the maximum frequency It becomes a state where one emotion type that becomes can not be specified.

  As described above, when it is not possible to specify one emotion type having the maximum emotion level or maximum frequency from the acquired emotion information, the above (2) is performed after the processing of (1). The processing of any of (7) to (7) may be executed, or the processing of assigning an individual emotion type for each unit by dividing into words, phrases, punctuation marks, etc. in the sentence. You may do it. These processes may be configured such that the user can switch the emotion imparting mode, or may be configured to automatically switch depending on conditions.

  When the process of acquiring the emotion information of the sentence from the words in the sentence ends in S302, the process executed by the information processing means 980 proceeds to the next determination of S304 "No emotion information in the sentence or normal [F0]?" .

  In S304, the information processing means 980 determines whether one emotion type could not be specified in S302, or whether the emotion information of the acquired text was information representing an emotion of “ordinary [F0]”. It is carried out.

  If emotion information other than “normal [F0]” has been acquired (S304; No), the process proceeds to “return” to S306, and proceeds to the process of S214 illustrated in FIG.

  If one emotion type could not be specified, or the emotion information of the acquired text is information representing an emotion of “normal [F0]” (S304; Yes), the information processing means 980 The process to be executed proceeds to the determination in S308 “Is the sentence at the beginning of a paragraph?

  In S308, the information processing means 980 determines whether or not the sentence being analyzed is the first sentence of the paragraph. If the sentence being analyzed is the first sentence of the paragraph (S308; Yes), the process proceeds to S310 “Set emotion information of sentence to“ Normal ”[F0]”. Processing is performed to generate a sentence with emotion associated with emotion information of “normal” [F0], and the process proceeds to S306.

  In S308, when it is determined that the sentence being analyzed is not the first sentence in the paragraph (S308; No), the process proceeds to S312 “Set same emotion information as previous sentence”, and the emotion of the previous sentence Processing for generating a sentence with emotion in which information is associated as emotion information of the sentence is performed, and the process proceeds to S306.

  Next, another embodiment of emotion addition processing when reading a sentence will be described with reference to FIG.

  FIG. 7 is a chart showing a process of generating a sentence with emotion from the input sentence.

  In the embodiment shown in FIG. 7, the user designates the reading of the story “Snow White”.

  When the user designates a text to be read out, the emotion adding device 900 inputs the text of the designated story from the text storage unit 20 (the process of S204 shown in FIG. 5). As a result of performing a single sentence analysis on the text of the input story (the process of S206 shown in FIG. 5), it is assumed that the text “Mirror mirror, what is the most beautiful in the world?” Is acquired (Example) 1 sentence).

  After the morphological analysis unit 28 performs processing for dividing the word of the sentence of Example 1 for each word, the emotion imparting unit 30 refers to the emotion dictionary 22 to acquire emotion information for each word, and the emotion information and the word are combined. When the associated sentence is generated, “mirror (noun) / yo (particle) | mirror (noun) /, (symbol) | this world (noun) / de (particle) | first (adverb [+2]) | The adjective [C1]) / I [F0] / noha (particle) | daare (noun) /? (Symbol) / "is generated (the processing of S208 and S210 shown in FIG. 5).

  Next, the emotion information associated with each word in the sentence of Example 1 is added to the above (2) by adding the emotion level of each word in the sentence for each emotion information to obtain the maximum emotion level. The processing is totaled using “a process of acquiring the emotion level and the added emotion level as the emotion information of the sentence”. Here, the sentence emotion acquisition means is based on the emotion emphasis information [+2] of the word “most”, the emotion type and emotion level [C1] (see FIG. 2A) of the word “beauty”. Then, emotion information of [C3] (Pleasure: Emotion Level 3) is acquired. Then, the emotion-added text generation unit 32 generates an emotion-added text that associates the text of Example 1 with the emotion information of [C3] (the process of S212 shown in FIG. 5).

  Next, the emotion-added sentence generating means 32 refers to the acoustic dictionary, acquires phoneme information corresponding to the emotion information of each word for each pronunciation unit, and generates a phoneme-added sentence in which the phoneme information is associated with each pronunciation unit. (Step S214 shown in FIG. 5). Then, the voice information generating means generates voice information for each pronunciation unit based on the phoneme information associated with the phoneme-added sentence, outputs the voice information to the output voice processing means 966, and the voice of the speaker or the like. The voice with emotion expression is uttered from the output means.

  In the case of the sentence of the first embodiment, the user can listen to the voice with emotion added by reading “Kagamiyo | Kagami | Konoyode / Ichiban / Utsukushinowa / Dare”. When the sentence of Example 1 is not the last sentence of the input story, the process of reading out the sentence of Example 2 is performed (the process of S216 shown in FIG. 5).

  Next, it is assumed that a sentence “Take that child into the forest and kill it” is acquired (referred to as a sentence of Example 2).

  After the morphological analysis unit 28 performs processing for dividing the word of the sentence of Example 2 for each word, the emotion imparting unit 30 refers to the emotion dictionary 22 to acquire emotion information for each word, and the emotion information and the word are combined. When the associated sentence is generated, “Anoko (Noun) / O (Participant) / Mori (Noun) / No (Participant) / Medium (Noun) / To (Participant) / Take (Verb) / ) / Killing (verb [B2] / [E2]) / do (ending) / deleting (auxiliary verb [B2]) / ”(processing of S208 and S210 shown in FIG. 5). .

  Next, the emotion information associated with each word in the sentence of Example 2 is added to the above (2) by adding the emotion level of each word in the sentence for each piece of emotion information, and obtaining the maximum emotion level. The processing is totaled using “a process of acquiring the emotion level and the added emotion level as the emotion information of the sentence”. Here, the sentence emotion acquisition means includes the emotion type and emotion level [B2] or [E2] (see FIG. 2 (a)) of the word “kill” and the emotion type and emotion level of the word “Otemai”. Based on [B2] (see FIG. 3), emotion information of [B4] (dynamic anger: emotion level 4) is acquired. Then, the emotion-added sentence generating unit 32 generates an emotion-added sentence that associates the sentence of Example 2 with the emotion information of [B4] (the process of S212 shown in FIG. 5).

  Next, the emotion-added sentence generating means 32 refers to the acoustic dictionary, acquires phoneme information corresponding to the emotion information of each word for each pronunciation unit, and generates a phoneme-added sentence in which the phoneme information is associated with each pronunciation unit. (Step S214 shown in FIG. 5). Then, the voice information generating means generates voice information for each pronunciation unit based on the phoneme information associated with the phoneme-added sentence, outputs the voice information to the output voice processing means 966, and the voice of the speaker or the like. The voice with emotion expression is uttered from the output means.

  In the case of the sentence of the second embodiment, a voice with added emotion can be uttered and read by the user by reading “Anocowo / Morino / Nakae / Tureteite / Colosseoshimai”. When the sentence of Example 2 is not the last sentence of the input story, the process of reading out the sentence of Example 3 is performed (the process of S216 shown in FIG. 5).

  Next, it is assumed that a sentence “What do you think?” Is acquired (referred to as a sentence of Example 3).

  After the morphological analysis unit 28 performs processing for dividing the sentence of the third embodiment for each word, the emotion imparting unit 30 refers to the emotion dictionary 22 to acquire emotion information for each word, and the emotion information and the word are combined. When an associated sentence is generated, “what (noun [+1]) / da (auxiliary verb) / to (particle) / thought (verb) / tsu (ending) / teruda (auxiliary verb) [* B2 / A1] / Is generated (the processing of S208 and S210 shown in FIG. 5).

  Next, the emotion information associated with each word in the sentence of Example 3 is added to the above (2) by adding the emotion level of each word in the sentence for each piece of emotion information, and obtaining the maximum emotion level. The processing is totaled using “a process of acquiring the emotion level and the added emotion level as the emotion information of the sentence”. Here, the sentence emotion acquisition means includes the emotion emphasis information [+1] of the word “what” (see FIG. 2A) and the emotion type and the emotion level [* B2] of the word ending with “Terunda”. Alternatively, emotion information of [B3] (dynamic anger: emotion level 3) is acquired based on [A1] (see FIG. 3). Then, the emotion-added sentence generating unit 32 generates an emotion-added sentence that associates the sentence of Example 3 with the emotion information of [B3] (the process of S212 shown in FIG. 5).

  Next, the emotion-added sentence generating means 32 refers to the acoustic dictionary, acquires phoneme information corresponding to the emotion information of each word for each pronunciation unit, and generates a phoneme-added sentence in which the phoneme information is associated with each pronunciation unit. (Step S214 shown in FIG. 5). Then, the voice information generating means generates voice information for each pronunciation unit based on the phoneme information associated with the phoneme-added sentence, outputs the voice information to the output voice processing means 966, and the voice of the speaker or the like. The voice with emotion expression is uttered from the output means.

  In the case of the sentence of the third embodiment, the user can listen to the voice with emotion added by reading “Nandato / Omotterunda”. If the sentence of the third embodiment is not the last sentence of the input story, the next sentence is read out, and if the sentence of the third embodiment is the last sentence, the emotion addition process is terminated (see FIG. Step S222 shown in FIG.

  Next, another embodiment of emotion addition processing when reading a sentence will be described with reference to FIG.

  FIG. 8 is a chart showing a process of generating a sentence with emotion information from the input sentence.

  In the embodiment shown in FIG. 8, the user designates the reading of the novel “Sherlock Holmes”.

  When the user designates a text to be read out, the emotion adding device 900 inputs the text of the designated story from the text storage unit 20 (the process of S204 shown in FIG. 5). Then, it is assumed that a sentence “It is almost time” has been acquired (referred to as a sentence of Example 4).

  After the morphological analysis unit 28 performs processing for dividing the word of the sentence of Example 4 for each word, the emotion imparting unit 30 refers to the emotion dictionary 22 to acquire emotion information for each word, and the emotion information and the word are combined. When the associated sentence is generated, a sentence “another (adverb) / soon (adverb) / dana (particle) /. (Symbol) /” is generated (the processing of S208 and S210 shown in FIG. 5).

  Next, the emotion information associated with each word in the sentence of Example 4 is added to the above (2) by adding the emotion level of each word in the sentence for each piece of emotion information. The processing is totaled using “a process of acquiring the emotion level and the added emotion level as the emotion information of the sentence”. However, emotion information is not associated with any of the words “already”, “soon”, “dana”, and “.” In the sentence of Example 4 (FIGS. 2A, 2B, and 2). (See FIG. 3). If the sentence of Example 4 is the sentence at the beginning of the paragraph, the sentence with emotion that associates the sentence of Example 4 with emotion information of [F0] (normal) is generated (S304 to S310 shown in FIG. 6). Processing.).

  Next, the emotion-added sentence generating means 32 refers to the acoustic dictionary, acquires phoneme information corresponding to the emotion information of each word for each pronunciation unit, and generates a phoneme-added sentence in which the phoneme information is associated with each pronunciation unit. (Step S214 shown in FIG. 5). Then, the voice information generating means generates voice information for each pronunciation unit based on the phoneme information associated with the phoneme-added sentence, outputs the voice information to the output voice processing means 966, and the voice of the speaker or the like. The voice with emotion expression is uttered from the output means.

  In the case of the sentence of the fourth embodiment, the voice is uttered by a normal reading which does not add feelings specially by reading “Moo / Solo Solodana” and the user listens to it. When the sentence of Example 4 is not the last sentence of the inputted story, the process of reading the sentence of Example 5 is performed (the process of S216 shown in FIG. 5).

  Next, it is assumed that the sentence “If you look at that ridiculous eye or a messy face” is acquired (referred to as the sentence of Example 5).

  After the morphological analysis means 28 performs processing for dividing the words of this Example 5 for each word, the emotion imparting means 30 refers to the emotion dictionary 22 to acquire emotion information for each word, and the emotion information and the word are combined. When the associated sentence is generated, “Ano (communicative) / Ozomashi (adjective ([E3] particle) / I (ending) / eye (noun) / ya (particle) / fair (adjective verb [E3]) / na ( (End of word) / Topper (adverb) / Shi (verb) / Ta (auxiliary verb) / Face (noun) / (Participant) / See (verb) / Ru (End of word) / and (Participant) /. A sentence is generated (the processing of S208 and S210 shown in FIG. 5).

  Next, the emotion information associated with each word in the sentence of Example 5 is added to the above (2) by adding the emotion level of each word in the sentence for each emotion information to obtain the maximum emotion level. The processing is totaled using “a process of acquiring the emotion level and the added emotion level as the emotion information of the sentence”. Here, the sentence emotion acquisition means has the emotion type and emotion level [E3] (see FIG. 2 (a)) of the word “Ozomashi”, and the emotion type and emotion level [E3] (see FIG. 2). 2 (a)), the emotion type [E6] and the emotion level are acquired. As shown in FIG. 4, since the maximum emotion level of emotion information [E] representing dynamic anger is set to “4” in relation to the acoustic dictionary 24, the sentence emotion acquisition means Then, a process of changing the acquired emotion information of [E6] to [E4] (static anger: emotion level 4) is performed. Then, the emotion-added sentence generating means generates an emotion-added sentence that associates the sentence of Example 5 with the emotion information of [E4] (the process of S212 shown in FIG. 5).

  Next, the emotion-added sentence generating means 32 refers to the acoustic dictionary, acquires phoneme information corresponding to the emotion information of each word for each pronunciation unit, and generates a phoneme-added sentence in which the phoneme information is associated with each pronunciation unit. (Step S214 shown in FIG. 5). Then, the voice information generating means generates voice information for each pronunciation unit based on the phoneme information associated with the phoneme-added sentence, outputs the voice information to the output voice processing means 966, and the voice of the speaker or the like. The voice with emotion expression is uttered from the output means.

  In the case of the sentence of the fifth embodiment, a voice with emotion added can be uttered by reading “Ano / Ozomasy / Meya / Osoroshigena / Noppelisita / Kaowo / Mirto” and can be heard by the user. If the sentence of the fifth embodiment is not the last sentence of the input story, a process of reading the next sentence of the sixth embodiment is performed (the process of S216 shown in FIG. 5).

  Next, it is assumed that the sentence “I feel like that when I look at Milverton” is acquired (referred to as the sentence of Example 6).

  After the morphological analysis unit 28 performs processing for dividing the sentence of Example 6 for each word, the emotion imparting unit 30 refers to the emotion dictionary 22 to acquire emotion information for each word, and the emotion information and the word are combined. When the associated sentence is generated, “Milverton (noun) / O (participant) / see (verb) / ru (ending) / to (particle) / just (adverb) / such (combined) / mood (noun) / (Sentence particles) / na (verb) / ru (ending) / danyo (auxiliary verb) /. (Symbol) / ”is generated (the processing of S208 and S210 shown in FIG. 5).

  Next, the emotion information associated with each word in the sentence of Example 6 is added to the above (2) by adding the emotion level of each word in the sentence for each piece of emotion information. The processing is totaled using “a process of acquiring the emotion level and the added emotion level as the emotion information of the sentence”. However, emotion information is not associated with any of the words in the sentence of Example 6 (see FIGS. 2A, 2B, and 3).

  If the sentence of Example 6 is the sentence at the beginning of a paragraph, a sentence with emotion that associates the sentence of Example 6 with emotion information of [F0] (normal) is generated (S304 to S304 shown in FIG. 6). Process of S310.) However, since the sentence of the fifth example exists before the sentence of the sixth example, the process proceeds to the process of S312 to add emotions associated with the same emotion information as the emotion information [E4] of the sentence of the previous fifth example. Generate sentences.

  Next, the emotion-added sentence generating means 32 refers to the acoustic dictionary, acquires phoneme information corresponding to the emotion information of each word for each pronunciation unit, and generates a phoneme-added sentence in which the phoneme information is associated with each pronunciation unit. (Step S214 shown in FIG. 5). Then, the voice information generating means generates voice information for each pronunciation unit based on the phoneme information associated with the phoneme-added sentence, outputs the voice information to the output voice processing means 966, and the voice of the speaker or the like. The voice with emotion expression is uttered from the output means.

  In the case of the sentence of the fourth embodiment, the voice is uttered by a normal reading that does not add special emotions in a reading of “Milvertonwo / Milt / Chodo / Sonna / Kibunni / Narudayo”, and the user listens. If the sentence of the sixth embodiment is not the last sentence of the input story, the next sentence is read out, and if the sentence of the sixth embodiment is the last sentence, the emotion addition process is terminated (see FIG. S222 shown in FIG.

  In this way, it is possible to read a sentence with a voice corresponding to the contents of the sentence.

  Next, if each emotion information associated with an independent word in the sentence cannot be used to identify the maximum emotion level or one emotion type that has the maximum frequency, the adjunct word in the sentence An embodiment in which the emotion information is acquired as the emotion information of the sentence will be described.

  When the user designates the reading of the sentence “I am a devil” (which is the sentence of Example 7), the morpheme analysis means 28 performs a process of dividing into words, and the emotion giving means 30 is the emotion dictionary. “I (pronoun) / ha (correspondent particle) / devil (noun [E2]) / de (connecting particle) / Yes” The sentence “Mass (Conclusive auxiliary verb [C1]) /. (Symbol) /” is generated. (Processing of S208 and S210 shown in FIG. 5).

  Next, the emotion information associated with each word in the sentence of Example 7 is used as the above-mentioned (4) “Process for obtaining the emotion information of the sentence using the emotion information of the attached word in the sentence”. Tally. In the text of Example 7, the word “devil” exists and is associated with [E2] representing the discomfort as the emotion type and emotion level. The emotion information of [C1] indicating the politeness associated with the word “is acquired, and the sentence with emotion is generated.

  Next, the emotion-added sentence generating means 32 refers to the acoustic dictionary, acquires phoneme information corresponding to the emotion information of each word for each pronunciation unit, and generates a phoneme-added sentence in which the phoneme information is associated with each pronunciation unit. (Step S214 shown in FIG. 5). Then, the voice information generating means generates voice information for each pronunciation unit based on the phoneme information associated with the phoneme-added sentence, outputs the voice information to the output voice processing means 966, and the voice of the speaker or the like. The voice with emotion expression is uttered from the output means.

  In the case of the sentence of the seventh embodiment, the voice is uttered with a polite reading of “Watashi / Ha / Akuma / De / Gozaimasu”. Thus, even if various emotion information is mixed in the sentence, it is possible to obtain appropriate emotion by acquiring emotion information of the ending word that is likely to express the emotion of the entire sentence. The sentence expressed by can be generated.

It is a figure which shows the signal processing type | system | group block diagram of the emotion addition apparatus which concerns on embodiment of this invention. It is a graph which shows the structure of the independent word emotion dictionary which linked | related the word of the independent word and emotion information. It is a graph which shows the structure of the independent word emotion dictionary which linked | related the word of the independent word and emotion information. It is a graph which shows the structure of the emotion dictionary of the adjunct word which linked | related the word of the adjunct word, and emotion information. It is a chart which shows the legend of the symbol used as emotion information. It is a flowchart of the process which automatically generate | occur | produces the audio | voice which added the emotional expression with respect to the input sentence. It is a flowchart of the process which acquires emotion information for every sentence. It is a graph which shows the process of producing | generating the sentence with an emotion from the input sentence. It is a graph which shows the process of producing | generating the sentence with an emotion from the input sentence.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Internet 20 Text memory | storage means 22 Emotion dictionary 24 Acoustic dictionary 26 Text input means 28 Morphological analysis means 30 Emotion provision means 32 Emotion sentence generation means 900 Emotion addition apparatus 960 Antenna 961 Wireless transmission / reception means 962 Wire transmission / reception means 970 Input means 971 Input interface 972 Display means 973 Display interface 980 Information processing means 981 RAM
983 ROM
984 Storage unit 990 Timer 965 Audio output unit 966 Output audio processing unit 999 Bus

Claims (16)

A sentence input means for inputting sentences;
Morphological analysis means for dividing the inputted sentence into words,
By referring to an emotion dictionary in which emotion information is associated with each word in advance, emotion information is acquired for each word separated from the input sentence, and emotion information is included in the sentence based on the acquired emotion information. Emotion imparting means for imparting,
Emotion-added text generating means for generating emotion-added text with emotion information given by the emotion giving means;
An emotion adding device characterized by comprising: The emotion giving means refers to an emotion dictionary that associates an emotion type and its emotion level as the emotion information for each word, acquires an emotion type and its emotion level for each word of the input sentence, The emotion adding apparatus according to claim 1, wherein emotion information is added to a sentence based on the emotion type, the emotion level, and the word. The feeling imparting means is
The emotion adding device according to claim 1, wherein the emotion adding device is provided for each specific unit in a sentence such as an independent word in a sentence, a clause, a bracket, or between punctuation marks. The emotion giving means refers to an emotion dictionary that associates an emotion type and its emotion level as the emotion information for each word, acquires an emotion type and its emotion level for each word of the input sentence, 4. The emotion adding apparatus according to claim 3, wherein emotion information is added to a sentence based on emotion information having a maximum emotion level among the acquired emotion types. The emotion imparting means adds emotion level values for each emotion type based on the emotion information acquired for each word, and among the added emotion types, the emotion type with the maximum emotion level value and the emotion type The emotion adding apparatus according to claim 3, wherein the added emotion level is added to the sentence. The feeling imparting means is
Refer to an emotion dictionary that associates words and emotion emphasis information in advance, acquire emotion emphasis information for each word of the input sentence, add or multiply the acquired emotion emphasis information to the emotion level, The emotion adding device according to claim 4 or 5, wherein emotion information is given in a sentence.   6. The emotion adding apparatus according to claim 5, wherein the word stored in association with the emotion emphasis information is a combination of a specific word and a word used as a pair with the specific word. The emotion giving means refers to an emotion dictionary that associates words of independent words and emotion information, and words of accessory words and emotion information, acquires emotion information for each word of the input sentence, 8. The emotion information is given to the sentence giving priority to the emotion of the attached word when there is different emotion information in the sentence based on the acquired emotion information. The emotion adding device according to item 1. The emotion imparting means adds the appearance frequency for each emotion type appearing in the sentence based on the emotion information acquired for each word, and selects the emotion type having the maximum appearance frequency among the added emotion types. The emotion adding device according to claim 3, wherein the emotion adding device is acquired as emotion information of a sentence. In the case where it is not possible to specify one emotion type having the maximum emotion level or maximum frequency from the acquired emotion information, the emotion imparting means writes the emotion type in the order of appearance. The emotion adding device according to claim 4 or 9, wherein the emotion adding device is provided inside. In the case where it is not possible to specify the maximum emotion level or one emotion type having the maximum frequency from each of the acquired emotion information, the emotion giving means sets the priority order set in advance for the emotion type. The emotion adding apparatus according to claim 4 or 9, wherein the emotion type having a higher priority among the plurality of acquired emotion information is used in the sentence. In the case where it is not possible to specify the emotion level that is the maximum emotion level or the maximum frequency from each of the acquired emotion information, the emotion imparting means, the independent words, phrases, punctuation marks in the sentence The emotion adding device according to claim 4 or 9, wherein the emotion adding device is divided by a unit such as an interval to give different emotion types. The sentence input means inputs a plurality of sentences,
In the case where it is not possible to specify the emotion level that is the maximum emotion level or the maximum frequency from each of the acquired emotion information, and there is a previous sentence before the sentence, The emotion adding device according to claim 4 or 9, wherein the emotion imparting means imparts emotion information of the previous sentence to the sentence.   Emotional sentence generation means, for each pronunciation unit such as phonemes, syllables, mora, morphemes, words, accent phrases, etc., the pitch or vowel formant of the voice when generating the voice, the voice strength, the voice speed, the voice intensity, Or, by referring to an acoustic dictionary that associates phoneme information such as intonation and emotion information in advance, phoneme information corresponding to the emotion information given by the emotion information giving means is acquired for each pronunciation unit, The emotion adding device according to any one of claims 3 to 13, wherein information is generated. Sentence input means for inputting a sentence, morphological analysis means for dividing a document into words, an emotion dictionary in which emotion information is associated with each word in advance, and emotion information is added to the sentence by referring to the emotion dictionary An emotion addition method using an emotion imparting means for generating an emotion-added sentence generating means for generating an emotion-added sentence,
Inputting a sentence from the sentence input means;
The morpheme analyzing means divides the input document into words;
The emotion giving means giving emotion information in a sentence with reference to the emotion dictionary;
The step of generating a sentence with emotion with the emotion information given by the feeling giving means;
An emotion addition method characterized by including: Sentence input means for inputting sentences, morphological analysis means for dividing the document into words, emotion giving means for giving emotion information in the sentences, sentiment-added sentence generating means for generating sentimented sentences, and the sentence input means An emotion addition program executed by the information processing means of an emotion addition apparatus comprising the information processing means for controlling the morpheme analysis means, the emotion giving means, and the emotion-added sentence generating means,
The information processing means is
A function of inputting a sentence from the sentence input means;
A function of separating the input document into words for the morphological analysis means;
The emotion imparting means refers to an emotion dictionary stored in advance by associating emotion information for each word, acquires emotion information for each word, and emotions in a sentence based on the acquired emotion information The ability to give information,
A function for generating a text with emotion with emotion information given by the emotion giving means to the text with emotion generation means;
Emotion addition program characterized by realizing.

Images