JP2006145891A - Voice processor, voice processing method, voice processing program and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a voice processor which smoothly copes with the voice uttered by a user. <P>SOLUTION: The voice processor is provided with: a voice recognition section 3 which recognizes uttered voice and converts the voice to character data Sc; a character data generating means 6 which determines whether recognized operation contents that are indicated by the character data Sc, match with the registered operation contents that are already registered in a database 8, and when the recognized operation contents do not match with the registered operation contents, generates related character data Sg that indicate operation contents related to the recognized operation contents, while acting as an artificial brain; and a voice synthesizing section 9 which responds to the voice by using the related character data Sg. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present application relates to a voice processing apparatus that executes voice response and operation corresponding to spoken voice.

  Conventionally, audio processing apparatuses have been used in various fields including car navigation. Here, specifically, the voice processing device is a device that provides various information corresponding to the voice spoken by the user (in the case of car navigation, a driver or a passenger) to the user by voice ( Patent Document 1).

Further, in recent speech processing devices, the speech recognition rate in the case of the word recognition method is a high level to some extent, so if the speech uttered by the user is a word previously registered in the speech processing device, Information corresponding to the spoken voice can be accurately provided to the user.
JP 2003-241797 A

  However, in the above-described conventional speech processing apparatus, since the content is not recognized unless the user speaks a word registered in the speech processing apparatus in advance, the user must always speak a fixed phrase. Therefore, there is a problem that voice processing using terms that are free for the user cannot be performed. In addition, even if recognized, only a response in a fixed sentence is made as a corresponding response. Therefore, only the same result is always output as a conversation, and as a result, the diversity is lost and the user uses the voice processing device. There was a problem of getting tired of using it.

  Furthermore, even if speech recognition is at a high level, if the user utters a word that is not registered in the speech processing device, the speech processing device cannot recognize the word, and the speech processing device becomes unresponsive. As a result, there was a problem that a feeling of resistance and refusal to voice operation was held.

  This application is made in view of such a problem, and makes it an example of providing the audio processing apparatus which can respond | correspond smoothly with the audio | voice of the free content which the user uttered.

In order to solve the above-mentioned problem, the invention according to claim 1 recognizes spoken voice and converts it into character data, storage means for storing pre-registered registration operation contents,
A character that generates related character data that represents an operation content related to the recognition operation content when the recognition operation content that is the operation content indicated by the character data does not match the registered operation content stored in the storage unit. Data generation means; and response means for responding to the voice using the generated related character data.

  In order to solve the above-mentioned problem, the invention according to claim 13 recognizes spoken speech and converts it into character data, a storage step of storing pre-registered registration operation contents, Character data for generating related character data indicating the operation content related to the recognition operation content when the operation content indicated by the character data does not match the registered operation content stored in the storage means A generating step, and a response step of responding to the voice using the generated related character data.

  In order to solve the above problem, an invention according to a fourteenth aspect causes a computer to function as the voice processing apparatus according to any one of the first to twelfth aspects.

  In order to solve the above-mentioned problem, in the invention described in claim 15, the voice processing program described in claim 14 is recorded so as to be readable by the computer.

  Next, the speech processing apparatus of the present application will be specifically described with reference to the drawings. In addition, embodiment described below is embodiment at the time of applying the audio processing apparatus of this application with respect to the car navigation system with a sound reproduction function.

(I) Embodiment FIG. 1 is a block diagram showing a schematic configuration of a speech processing apparatus of the present application.

  As shown in FIG. 1, the speech processing apparatus V according to the embodiment includes a microphone 1, an A / D conversion unit 2, a speech recognition unit 3 as speech recognition means, a first operation execution unit, and a second operation execution. An operation execution unit 4 as a means, a control unit 5 as an execution process determination unit and a character data generation process determination unit, a character data generation unit 6 as a character data generation unit, an analysis unit 7, and a database as a storage unit 8, a first voice synthesis unit, a voice synthesis unit 9 as a second voice synthesis unit, a D / A conversion unit 10, and a speaker 11.

  Next, the overall operation will be described.

  In the above configuration, when a voice is spoken by the user, the voice signal Sa corresponding to the voice is output to the A / D conversion unit 2 through the microphone 1. Then, the audio signal Sa is converted into audio data Sb by the A / D conversion unit 2 and output to the audio recognition unit 3. Thereafter, the content of the voice data Sb is voice-recognized by the voice recognition unit 3 to be converted into corresponding character data Sc and output to the control unit 5.

  Thereafter, the control unit 5 recognizes the registered operation content (hereinafter simply referred to as “recognition operation content”) that is the operation content indicated by the character data Sc (hereinafter simply referred to as “recognition operation content”). It is determined whether or not the content matches the “operation content”. The details of the registration operation will be described later.

  Here, the operation content is the content of navigation processing that can be executed by the car navigation system according to the embodiment, for example, map display for navigation, route search or route guidance using the search result, or MD as a sound reproduction function. This refers to the playback processing of music recorded on (Mini Disc).

  If the recognition operation content matches the registration operation content, the operation execution unit 4 executes the registration operation content having the same content as the recognition operation content (hereinafter simply referred to as “operation execution processing”). Do).

  Further, the character data Sc indicating the registered operation content is output to the speech synthesizer 9, synthesized by the speech synthesizer 9, converted into speech data Sd, passed through the D / A converter 10, The sound is converted into Se, and the speaker 11 utters the sound corresponding to the sound signal Se.

  By the above operation, when the recognition operation content matches the registration operation content, the operation execution unit 4 executes the registration operation content having the same content as the recognition operation content obtained by the user's utterance. The intended operation is performed.

  On the other hand, when the recognition operation content that is the operation content indicated by the character data Sc does not match the registered operation content, the character data Sc is output from the control unit 5 to the character data generation unit 6.

  Here, the character data generation unit 6 generates character data indicating the operation content related to the recognition operation content (hereinafter referred to as “related character data”). It is realized by operating a conversation program as a brain.

  Artificial brainless is a general term for conversational programs that are used to confront general “artificial intelligence”. In the process of so-called bottom-up artificial intelligence, it is complicated to reach “humanity”. Contrary to this, it is a general term for conversation programs that try to create “humanity” by forming a model of “humanity” from the top down. The origin of this word was called “artificial incompetence” in contrast to “artificial intelligence”, which is useful for things that are not human, such as search engines such as the Internet and expert systems. As a result of hating negative images, in recent years it has come to be called “artificial brainless”.

  The related character data generated by the character data generation processing performed by the character data generation unit 6 includes, for example, any word in a word key string (which will be described later) of character data indicating the recognition operation content. Things that have the same readings, things that are semantically similar, things that have meanings connected, and related when the user recognizes words spoken by related character data based on the speech spoken by the user It means that the user can easily understand and understand that character data has been generated and responded.

  Therefore, the character data generation process performed by the character data generation unit 6 may generate, as related character data Sg, an unexpected word that is not connected to the character data Sc obtained by the user speaking. For example, a word centering on the character data Sc meaning the registered operation content may be generated as the related character data Sg.

  Thereby, even if the recognition operation content obtained by the user's utterance does not match the registered operation content, the voice processing device V can cause the user to make some response.

  Next, the operation content of the character data Sc output by the character data generation unit 6 is decomposed into parts of speech by the analysis unit 7, converted into character data Sf, and stored in the database 8. Thereafter, a plurality of word keys Si that are related to the character data Sf obtained as a result of the part-of-speech decomposition and mean the registered operation contents are output from the database 8 to the character data generation unit 6. The character data generation unit 6 generates related character data Sg based on the word key Si and outputs it to the control unit 5.

  Thereafter, the action execution unit 4 obtains the action contents indicated in the related character data Sg, specifically, the action contents indicated in the related character data Sg as described above, that is, the user speaks. The action content generated as the related character data Sg is an unexpected word that has no meaning to the character data Sc, or the word centered on the character data Sc meaning the registered operation content is generated as the related character data Sg. The operation content and the like are executed.

  Further, the character data Sg is output to the voice synthesizer 9 to be synthesized, converted into voice data Sk, converted into a voice signal Sl by the D / A converter 10, and corresponds to the voice signal S1 by the speaker 11. Voice is spoken.

  By this operation, even if the recognition operation content obtained by the user's utterance does not match the registered operation content, the character data generation unit 6 generates the related character data Sg, and the voice processing device V causes the user to do something. By making a response, it is possible for the user to interact with a free expression without speaking a pre-registered fixed sentence, and to avoid the voice processing device V from becoming unresponsive. Can reduce the sense of resistance and refusal to voice operation.

  Next, the audio processing of the audio processing apparatus will be specifically described with reference to FIGS.

  First, FIG. 2 is a flowchart showing the entire sound processing of the sound processing apparatus.

  As voice processing according to the embodiment, first, voice uttered by a user is voice-recognized by a voice recognition unit and converted into character data Sc (step S21).

  Then, it is determined whether or not the recognition operation content, which is the operation content indicated by the character data Sc, matches the registered operation content (step S22).

  Here, regarding step S22, an example of registration operation content will be described with reference to Table 1.

当該登録動作内容としては、例えば表１に示すように、「ＭＤ再生」、「ナビ画面表示」、「次の曲を再生」、「前の曲を再生」又は「渋滞情報確認」等の、動作実行部４において実行可能な実施形態に係るナビゲーションシステムとしての具体的な動作内容が、予め登録されている。そして、ステップＳ２２では、これらの登録動作内容と認識動作内容とが一致しているか否かを判断する。

As the registration operation content, as shown in Table 1, for example, “MD playback”, “navigation screen display”, “play next song”, “play previous song”, or “congestion information confirmation”, etc. Specific operation contents as the navigation system according to the embodiment that can be executed by the operation execution unit 4 are registered in advance. In step S22, it is determined whether or not the registered operation content matches the recognized operation content.

  If the recognition operation content, which is the operation content indicated by the character data Sc, does not match the registration operation content (step S22; NO), it is next stored in the database 8 as a part of the registration operation content. It is determined whether or not the recognition operation content includes a partial operation registered in advance or a partial content (hereinafter simply referred to as “partial content”) that is a content attached to the operation (step S23).

  Here, the example of the partial content about step S23 is demonstrated using Table 2. FIG.

ステップＳ２３についての部分内容としては、例えば表２左に示すように、「ＭＤ」、「ナビ」、「次」、「前」、「天気」等が、予め登録されている。

For example, “MD”, “Navi”, “Next”, “Previous”, “Weather”, etc. are registered in advance as the partial contents of Step S23.

  In step S23, for example, when the user utters a voice saying “The weather has been bad recently”, the content of the character data Sc obtained from the voice uttered by the user is analyzed by the analysis unit 7, It is determined whether or not the content as the analysis result matches any of the partial contents as the process of step S23. For example, when “weather” which is the partial content shown in Table 2 is included in the recognition operation content, the recognition operation content includes the partial content (that is, a part of the operation content indicated by the character data Sc). Is the same as the partial content in step S23) (step S23; YES). Similarly, if the user utters a voice saying “I bought a new MD today”, the partial contents “MD” shown in Table 2 are included in the recognition operation contents, so the part about step S23 It is determined that the content is included (step S23; YES).

  On the other hand, if it is determined in S23 that no partial content is included (step S23; NO), specifically, if there is no partial content as shown in Table 2 in the recognition operation content, The data generation unit 6 generates related character data indicating the operation content related to the recognition operation content (step S28). Details of the generation processing of the related character data Sg will be described later.

  Next, in the determination in step S23, if the recognition operation content includes partial content (step S23; YES), whether or not to perform an operation execution process as a process for executing the registration operation content corresponding to the partial content is determined. Is determined (step S24). Specifically, as described above, when the recognition operation content “MD” uttered by the user includes partial content, it is determined whether or not to perform a corresponding operation execution process based on the partial content.

  Here, the criteria for determining whether or not to perform the operation execution process in step S24 will be described with reference to Table 3.

当該動作実行処理を行うかの判断基準としては、例えば、表３に示すように、ユーザが走行中であるか否か、又は他の操作の最中であるか否か等が判断要素となる。更に具体的には、ユーザが車を運転中である場合に、人工無脳としての文字データ生成部６により関連文字データＳｇを生成して応答した結果、それに伴って、例えば急激な音量変化又は強制的なルート変更等が発生した場合、車を運転中であるユーザを驚かせてしまう可能性がある。このような車走行中のユーザを驚かせる可能性があるような応答は回避する必要があるため、ユーザが車を運転中である場合には文字データ生成処理を行わないことが適切であると考えられる。よって、このような可能性がある場合は、人工無脳の機能を用いずに動作実行処理を行うと判断される（ステップＳ２４；ＹＥＳ）。

For example, as shown in Table 3, whether or not the user performs the operation execution process is determined based on whether or not the user is running or whether another operation is being performed. . More specifically, when the user is driving the vehicle, the character data generation unit 6 as an artificial brainless unit 6 generates the related character data Sg and responds. As a result, for example, a sudden volume change or When a forced route change or the like occurs, there is a possibility that the user who is driving the vehicle will be surprised. Since it is necessary to avoid such a response that may surprise the user while driving, it is appropriate not to perform the character data generation process when the user is driving the car. Conceivable. Therefore, when there is such a possibility, it is determined that the action execution process is performed without using the artificial brainless function (step S24; YES).

  On the other hand, if it is determined in step S24 that the action execution process is not performed (step S24; NO), the related character data Sg should be generated using the artificial brainless function as the character data generation unit 6. Then, the process proceeds to step S28 described later.

  Next, when it is determined in step S24 that the operation execution process is to be performed (step S24; YES), the operation execution process (the operation execution process corresponding to the partial contents) shown in Table 2 is performed. (Step S25). Specifically, the operation content shown on the right side of Table 2 as the registered operation content corresponding to the partial content on the left side of Table 2 is performed by the operation execution process. For example, when “MD” on the left side of Table 2 corresponds to “MD” on the right side, “Play MD” is executed in the action execution unit 4, and when “Navi” corresponds to “Navigation screen on the right side” “Display” is executed in the action execution unit 4. When corresponding to “next”, “play next song” on the right side is executed in the action execution unit 4, and “previous” In the case of corresponding to “play music” and “weather”, the operation execution unit 4 executes “display weather forecast screen” on the right side. Here, as described above, the case where “play MD” is selected as the registration operation content corresponding to “MD” is shown, but in addition to the registration operation content shown in Table 2, “MD” For example, in the case where “record MD”, “stop MD”, etc. are registered as the registered contents corresponding to, after selecting which of these registered operation contents is to be executed, Since the operation content is executed, the operation execution process in this case includes a process of selecting the operation execution content from the registered content corresponding to the partial content and executing the operation execution content.

  Next, it is determined whether or not to perform character data generation processing, which is processing by the character data generation unit, based on the following criteria (step S26).

  Here, with respect to step S26, the criteria for determining whether or not to perform the character data generation process will be described with reference to Table 4.

表４に示すように、ステップＳ２６の判定において文字データ生成処理を行うか否かを判断する基準として、例えば、ＭＤ再生等がなされた場合には、音楽が出力されればＭＤが再生されるという動作が行われたことがわかるので、このような場合には、動作内容を音声で確認するまでもない。また、例えばＭＤの再生、停止、音量調節等の頻度の高い動作がなされた場合であって、この動作内容が行われた際に「ＭＤを再生しました」、「ＭＤを停止しました」等を何度も言われたく無い場合も、動作内容を音声で確認しないことが望まれる。よって、このような場合には、文字データ生成処理を行う（ステップＳ２６；ＹＥＳ）。

As shown in Table 4, as a criterion for determining whether or not to perform the character data generation process in the determination in step S26, for example, when MD reproduction is performed, MD is reproduced if music is output. In such a case, it is not necessary to confirm the operation content by voice. Also, for example, when an operation with a high frequency such as MD playback, stop, volume control, etc. is performed, when this operation content is performed, “MD was played”, “MD was stopped”, etc. Even if you do not want to be told many times, it is desirable not to confirm the operation content by voice. Therefore, in such a case, a character data generation process is performed (step S26; YES).

  If the character data generation process is appropriate, the related character data Sg is generated. In the character data generation process in this case, an unexpected word that is not connected to the character data Sc obtained by the user's utterance may be generated as the related character data Sg, or the registered operation content is meant. The word centering on the character data to be generated may be generated as the related character data Sg (step S28). For example, if the character data meaning the registered operation content is “MD playback”, “I want to listen to MD but I don't have a good MD”. Furthermore, even if the word “go home” uttered by the user includes partial content, if the focus is on the fun of conversation without performing the action corresponding to the partial content, As a result of the data generation processing, related character data Sg indicating a word such as “dislike” is generated (step S28).

  In parallel with the character data generation process (step S28) for generating the related character data Sg described above, the normal operation as the car navigation system (that is, the operation on the character data Sc obtained by the user speaking) For example, the map display processing to the home as the operation corresponding to the instruction to return home is completed in the above-described step S25, and the processing in the above step S28 is only for enjoying the conversation. The related character data Sg generated by the character data generation process is used as the object.

  Next, the related character data Sg generated by the character data generation process is subjected to a voice synthesis process by the voice synthesizer 9, that is, a process of converting the related character data Sg into the voice data Sk. Then, a response is output to the user through the speaker 11 (steps S29 and S30). For example, if there is character data that means the registered operation content “confirmation of traffic information”, traffic information will be displayed on the screen of the car navigation etc., depending on the voice. Etc. "is made.

  On the other hand, in the determination in step S26, as the above-described determination criterion, for example, in order to avoid a response that may surprise the user who is running and to avoid danger, the processing result is more correctly than the fun of the conversation. When giving priority to the user (step S26; NO), the voice synthesizer 9 performs a response process by voice corresponding to the registration operation content performed previously (step S25), and the speaker 11 is used as voice. The response is output via (steps S29 and S30). Therefore, in this case, both the action performed by the action execution process of step S25 and the voice corresponding process to be voiced after step S27 do not involve the character data generation process and use the action execution process. Will be done.

  On the other hand, if the determination in step S23 or step S24 is NO (that is, the recognition operation content does not include even a part of the registration operation content (step S23; NO)), or the recognition operation content is one of the registration operation content. The process is not executed (step S24; NO), but the process proceeds to step S28. In this case, the action execution process (step S25) is performed. Since it is not executed, as in the case described above, the response is made by the related character data Sg generated by the character data generation unit 6 for both operation correspondence and voice correspondence.

  Next, when it is determined in step S22 that the recognized action content matches the character data Sc that means the registered action content (step S22; YES), the action content of the character data Sc is executed (step S22). S31). As described above, the operation content executed here is the registration operation content having the same content as the recognition operation content, and the content shown in the registration operation content is executed.

  Next, as the operation after step S26 described above, it is first determined whether or not to perform character data generation processing. That is, as described above, the registration operation content is executed by the operation execution processing having the same content as the operation on the character data Sc obtained by the user's utterance, and further by the related character data Sg generated by the character data generation processing. When it is appropriate that a response is made (step S26; YES), character data generation processing is performed (step S28), and when it is given priority to correctly convey the processing result to the user (step S26; NO). Is determined to be a case where the response by voice is performed together with the action content performed earlier, and the response by the character data meaning the registered operation content is performed, and in response to the determination, the response by voice is performed by the voice synthesis processing (step S29, S30).

  Next, the processing in step S28 of FIG. 2 (related character data Sg generation processing by character data generation processing) will be specifically described with reference to FIG. FIG. 3 is a flowchart showing related character data generation processing by character data generation processing used in the present application.

  First, as a premise for generating the related character data Sg by the character data generation process, a “word key” used for generating the related character data Sg is registered in the database 8.

  Here, as described in FIG. 1, the word key refers to a word that is related to the character data Sf obtained as a result of the part-of-speech decomposition of the recognition operation content and means the registered operation content. For example, to register the "Hello" as a word key Si. Specifically, this will be described with reference to Table 5.

表５に示すように、第１の単語キーＳｉとして、出現形である「こんにちは」と登録すると共に、当該出現形の読み「コンニチハ」、原形「こんにちは」、原形の読み「コンニチハ」、品詞「感動詞」等の単語キーを登録しておく。同様に、第２の単語キーＳｉとして、出現形である「初め」と登録し、出現形の読み「ハジメ」、原形「初め」、原形の読み「ハジメ」、品詞「名詞」等のデータを登録しておく。このように、第３の単語キーＳｉに「まして」、第４の単語キーＳｉに「お年玉」といった単語キーＳｉを登録する。ここで、単語キーＳｉは自ら登録してもよいし、予めデータベース８に登録されているものを関連文字データ生成に使用してもよい。

As shown in Table 5, as the first word key Si, the appearance shape and registers as "Hello", reading of the appearance form "Hello", original "Hello", reading original "Hello", the part of speech " Register a word key such as “adverb”. Similarly, as the second word key Si, “first” which is an appearance form is registered, and data such as “hajime” which is the appearance form, “hajime” which is the original form, “hajime” which is the original form, and part of speech “noun” are stored. Register. In this way, the word key Si such as “Noshi” is registered in the third word key Si and “Otoshi” is registered in the fourth word key Si. Here, the word key Si may be registered by itself, or a word key registered in advance in the database 8 may be used for generating related character data.

  Here, the counter shown in Table 5 is counted when the user uses the word key Si. The greater the number of counters, the higher the frequency of use by the user, and the higher the probability that the related character data Sg will be generated based on this information.

  Furthermore, it demonstrates concretely using Table 6. FIG.

表６は、上記表５で登録された単語キーＳｉがユーザによってどの程度の頻度でテーマとして使われたかを登録するため、かつてユーザが使用した単語キーの中から、名詞、動詞、感動詞を抜き出してそれぞれの単語キーＳｉが、テーマとして使われた順番と、単語キーを使用した時間（グリニッジ標準時からの経過時間）と共に登録される。

Table 6 shows how often the word key Si registered in Table 5 is used as a theme by the user. From the word keys used by the user, nouns, verbs, and emotional verbs are stored. Each extracted word key Si is registered together with the order in which it was used as a theme and the time for which the word key was used (elapsed time from Greenwich Mean Time).

  As a result, it can be seen that a user with a large theme number and an acquisition time close to the current time has been recently used by the user, so that the probability of becoming a theme when generating related character data in the future increases.

  Next, related character data generation by the related character data generation processing using the word key Si registered in the database 8 will be described.

  As shown in FIG. 3, it is set whether or not the voice processing device V responds stochastically after the voice is spoken by the user (step S41). Next, it is determined whether or not the result set by the user responds (step S42). If the setting result is not answered (step S42; NO), the operation is terminated (step S43). If it is determined that the setting result is a response (step S42; YES), then, a theme used for generating related character data is set by the user, thereby acquiring a theme used for generating related character data (step S44). In accordance with this theme, the related character data Sg is generated later.

  The themes here are nouns, verbs, emotional verbs and the like as described above. Using this word, the related character data Sg is generated based on the word. As described above, a theme for generating the related character data Sg is set from the word keys Si registered in the database 8 when replying.

  Then, after recognizing the voice uttered by the user, a response time indicating how long a response is made by the voice processing device V is recorded (step S45). Next, theme selection is performed (step S46). At this time, a theme having the same reading as the theme word is searched and selected at random. Thereafter, the word key string is acquired (step S47).

  Here, the details will be described with reference to Table 7.

表７は、表５に登録された単語キーＳｉに対する番号がそれぞれ第１単語、第２単語、第３単語の順で並んでおり、それを組み合わせて文字データとして表されることになる。

In Table 7, numbers for the word keys Si registered in Table 5 are arranged in the order of the first word, the second word, and the third word, respectively, and are combined and expressed as character data.

  As shown in Table 7, a sequence having 0 in the first word is the first clause. Conversely, if the third word has 0, that is the end clause. The arrangement of the word key 1, so are arranged in the order of 0-1-0, made to the string "Hello". Furthermore, since the word key sequence 2 is 0-2-3, it becomes a character string “Nice to meet you”.

  When the theme for generating the related character data Sg is “8 New Year” in Table 5 (when the theme data 4 or 5 is determined as the theme), the second word is “8”. call. That is, as shown in Table 7, a word key string whose word key sequence is “7-8-9” is called.

  Next, when the connected related character data is generated, it is determined whether there is a word key Si connected before the word key string (step S48). If there is a word key Si connected before the word key string (step S48; YES), the word key Si is added before the word key string (step S49). When the word key string is “7-8-9”, it is connected to “0-7-8” to become “0-7-8-9”. If there is a word key Si connected before the word key string until 0 comes to the beginning of the sequence (step S48; YES), the operation of adding the word key Si before the word key string (step S49) is performed. You can continue. If 0 comes to the beginning of the sequence, this operation is terminated, and then it is determined whether there is a word key Si connected after the word key string (step S50).

  If there is no word key Si connected before the word key string (step S48; NO), it is subsequently determined whether there is a word key Si connected after the word key string (step S50).

  Next, when there is no word key Si connected after the word key string (step S50; NO), the word key string is converted into a character string (step S52).

  If there is a word key Si connected after the word key string (step S50; YES), the word key Si is added before the word key string (step S51). When the word key string is “0-7-8-9”, it is connected to “8-9-10” to become “0-7-8-9-10”. Similarly, until there is a word key Si connected before the word key string until 0 comes to the end of the sequence (step S48; YES), the word key Si is added before the word key string (step S49). ) Is continued. When 0 comes to the end of the sequence, this operation is terminated, and then the word key string is converted into a character string (step S52).

  When the word key sequence “0-7-8-9-10-0” is determined, the appearance form of the word based on it is assigned.

  Since 0 is considered as a symbol of the beginning and end, and no actual word is assigned, the respective words corresponding to 7, 8, 9, and 10 are determined according to Table 5, and "Now, New Year-It is-" Is generated.

In this way, three words different from the central word are searched from the database 8 and connected to generate character data.
The more natural the connection of the three words registered in the database 8, the more meaningful the output sentence will be for humans.

  As described above, even if the recognition operation content obtained by the user's utterance does not match the registration operation content, the character data generation unit 6 generates the related character data Sg, and the voice processing device V informs the user. Although some response can be made, if the word key Si used for generating the related character data Sg cannot be updated, the freshness of the word key Si used for generating the related character data cannot be maintained. The fun of conversation is halved.

  Therefore, since fresh and interesting related character data is generated by the character data generation process, the word key Si used for generating the related character data Sg is updated.

  Hereinafter, a method of updating the word key Si stored in the database 8 used in the present application will be described with reference to FIG. FIG. 4 is a flowchart when the character data stored in the database used in the present application is updated.

  When updating the database 8, first, character data obtained by uttering by the user is analyzed by the analysis unit, the part of speech is decomposed, and divided into words (step S61). As shown in Table 7, it is determined whether or not the word key Si has an arrangement of three words (step S62). If there is no arrangement of three words (step S62; NO), no registration is made (step S66). If there is a sequence of three words (step S62; YES), it is next determined whether or not the sequence is registered in the database (step S63).

  If the sequence is registered in the database 8 (step S63; YES), it is not necessary to register again, so that it is not registered (step S66). If the line is not registered in the database 8 (step S63; NO), it is determined whether to register word data and word line data (step S64). And when not registering (step S64; NO), it is not registered (step S66), and when registering (step S64; YES), the word key Si or the word key of noun, verb, impression verb, etc. The line is registered (step S65).

  In this way, since the character data stored in the database 8 can be updated to be used for generating the related character data Sg, the freshness of the character data used when generating the related character data is maintained. Will be drunk.

  It is also conceivable to update the database 8 using a network.

  As described above, even if the user does not utter a standard sentence registered in the speech processing device, the character data generation means ignores the word uttered by the user based on the word or sometimes, Since a certain sentence is created, it is possible to talk using a free expression without speaking a fixed sentence.

  Therefore, since conversations always produce different results, there are diversity, and conversation can be enjoyed as a catch ball of words that do not know what happens, and an audio device that does not get bored can be realized.

  In addition, even if the user utters a word that is not registered in the speech processing device, the character data is generated and responded. The situation can be eliminated and the sense of resistance and refusal to voice operation can be reduced.

  Roughly, even if the voice is misrecognized, any response can be made if importance is given to the fun of conversation, so that the state of no response can be avoided.

  In the above-described embodiment, when the recognition operation content matches or does not match the registered operation content, if the emphasis is on the fun of conversation, both the action correspondence and the voice dialogue are both Both may perform character data generation processing. In this case, it is not determined from the beginning whether the character data matches the character data indicating the registered operation content.

  Furthermore, although the above-described embodiment has shown the form in which the response by the operation is performed first and the response by the voice is performed thereafter, the response by the voice may be performed first.

  The above-described embodiment is based on the assumption that the voice uttered by the user is correctly recognized. However, if the voice is recognized and converted into character data, it is already erroneously recognized. Even so, the present invention can be used effectively.

  Here, misrecognition was not correctly determined when determining whether the recognition operation content matches the registered operation content or when determining whether the recognition operation content contains partial content. Refers to cases. For example, when the recognition operation content matches the registration operation content, if it is determined that the recognition operation content does not match the registration operation content (S22; No), or the recognition operation content does not match the registration operation content. When it is determined that the recognition operation content does not match the registered operation content and the recognition operation content does not include the partial content (S22; No → S23; No) when the recognition operation content includes the partial content, Furthermore, when both the registered operation content and the partial content are not included, there is a case where it is determined that the registered content is not included, but the partial content is included (S22; No → S23; Yes). If there is a misrecognition, there may be cases where the user's truly intended response may not be made. Alternatively, since some response is made by the character data generation process, a state of no reaction can be avoided.

  Furthermore, as described above, in order to make a recognition error in order to enjoy a conversation, for example, if erroneous recognition is executed 10 times per minute, then correct recognition is performed, and character data generation processing is performed based on the recognition. It is also possible to make settings such as

  As described above, the audio processing device, the audio processing method, the audio processing program, and the recording medium have been described. However, not only audio but also a combination of video and audio can be similarly handled.

  For example, in the above-described embodiment, the response of the voice processing device to the voice uttered by the user has been shown. However, the response processing by the processing device using video for the voice uttered by the user can be applied. is there. In this case, the above-described processing using video is performed in S28 to S30 in FIG.

  Furthermore, as a modification of these, response processing by a processing device using video will be described.

(II) Modified Embodiment In the embodiment described above, the content of the voice input from the user is used as the recognition operation content, whereas in the following modified embodiment, the result of capturing the user's movement as a video and recognizing the content is used. Used as recognition operation content.

  More specifically, first, the user's movement is recognized by the camera, the movement is acquired as data, data having the same movement is searched from the database, and the read data is combined using an artificial brainless function. A new motion is generated and displayed on a display unit (not shown). In this case, sometimes it may be an action that imitates the movement completely, but when using artificial brain, it does not expect a uniform action from the beginning, so various actions will be taken regardless of the movement of the user It is possible to display and respond.

  Further, a program corresponding to the flowchart shown in FIG. 1 is recorded on a recording medium such as a flexible disk or a hard disk, or is acquired and recorded via a network such as the Internet, and is read by a microcomputer or the like. When executed, the microcomputer can function as a control unit according to each embodiment.

  Note that the sound processing device, sound processing method, sound processing program, and recording medium of the present application are not limited to the above-described embodiments, and are mounted on a television, an audio system, or the like, or sound that responds to free words. The present invention can also be used for devices and the like.

It is a block diagram which shows schematic structure of the audio processing apparatus of this application. It is a flowchart which shows the whole audio | voice process of the audio | voice processing apparatus of this application. It is a flowchart which shows the related character data generation by the character data generation process used for this application. It is a flowchart at the time of the character data preserve | saved at the database used for this application being updated.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Microphone 2 ... A / D conversion part 3 ... Voice recognition part 4 ... Operation | movement execution part 5 ... Control part 6 ... Character data generation part 7 ... Analysis part 8. .. Database 9 ... speech synthesizer 10 ... D / A converter 11 ... speaker

Claims (15)

Speech recognition means for recognizing spoken speech and converting it into character data;
Storage means for storing pre-registered registration operation content;
A character that generates related character data that represents an operation content related to the recognition operation content when the recognition operation content that is the operation content indicated by the character data does not match the registered operation content stored in the storage unit. Data generation means;
Response means for performing a response to the voice using the generated related character data;
An audio processing apparatus comprising: The speech processing apparatus according to claim 1,
The character data generating means does not match the registration operation content with the registration operation content, and any one of the partial contents stored in the storage means as a part of the registration operation content. The speech processing apparatus, wherein the related character data is generated when the recognition operation content is not included. The speech processing apparatus according to claim 2,
Execution to further determine whether or not to execute the registration operation content corresponding to the partial content when the recognition operation content does not match the registration operation content and the recognition operation content includes the partial content Further comprising a process determination means;
The speech processing apparatus, wherein the character data generation unit generates the related character data when the execution process determination unit determines that the registration operation content corresponding to the partial content is not to be executed. The voice processing device according to claim 3,
A speech processing apparatus, further comprising: a first operation executing unit that executes the registered operation content when the execution process determining unit determines to execute the registered operation content corresponding to the partial content. The speech processing apparatus according to claim 4,
The voice processing device, wherein the character data generation unit generates the related character data when the first operation execution unit executes the registration operation content corresponding to the partial content. The speech processing apparatus according to any one of claims 1 to 5,
The speech processing apparatus further comprising first speech synthesis means for converting the related character data generated by the character data generation means into speech data and generating a corresponding speech. The speech processing apparatus according to any one of claims 4 to 6,
The voice processing apparatus according to claim 1, wherein the first action executing means executes action contents of the related character data generated by the character data generating means. The speech processing apparatus according to any one of claims 1 to 7,
The speech processing apparatus further comprising: a second operation executing unit that executes the registered operation content when the recognized operation content matches the registered operation content. The speech processing apparatus according to claim 8, wherein
The voice processing apparatus, wherein the character data generation unit generates the related character data when the second operation execution unit executes the registered operation content. The speech processing apparatus according to claim 8, wherein
The voice processing apparatus, wherein the character data generation unit generates registration operation character data related to the registration operation content when the second operation execution unit executes the registration operation content. The speech processing apparatus according to any one of claims 8 to 10,
2. A speech processing apparatus, further comprising: second speech synthesis means for converting character data indicating the content of the operation executed by the second operation execution means into speech data and emitting a corresponding speech. The speech processing apparatus according to any one of claims 8 to 11,
The voice processing apparatus, wherein the second operation executing means executes the operation content of the related character data generated by the character data generating means. A speech recognition process for recognizing spoken speech and converting it into character data;
If the recognition operation content, which is the operation content indicated by the character data, does not match the registered operation content stored in advance in the storage unit, related character data that means the operation content related to the recognition operation content is generated. A character data generation process;
A response step of performing a response to the voice using the generated related character data;
An audio processing method comprising:   An audio processing program for causing a computer to function as the audio processing device according to any one of claims 1 to 12.   15. An information recording medium in which the audio processing program according to claim 14 is recorded so as to be readable by the computer.

Images