JP2023524374A - Method for providing voice recognition music selection service and voice recognition music selection device - Google Patents

Abstract

A method for providing a voice recognition music selection service is disclosed. According to one aspect of the present invention, a step of providing a plurality of reference tokens generated by segmenting a reference utterance containing a song title and a singer name, segmenting on the basis of parts of speech, and stems and endings, a user a step of inputting a search utterance sentence uttered by a to search for a song in the form of text; a step of dividing the search utterance sentence into words, a part of speech, and a step of generating a plurality of search tokens by dividing the word stem and ending into a plurality of search tokens; comparing the search token with a plurality of reference tokens; calculating a similarity score of the search utterance sentence to the reference utterance sentence according to the comparison result; and providing song information to the user based on the similarity score. A method for providing a voice recognition song selection service is provided.
[Selection drawing] Fig. 2

Description

The present invention relates to a voice recognition music selection service providing method and a voice recognition music selection device.

In general, in order to search for a song that a user wants to sing on a song accompaniment instrument, a search remote controller is used to press a number or letter dial provided on the remote controller to input the song name, singer name, or song number of the corresponding song. and so on have been used.

However, such a method has the problem that it takes a lot of time and effort to search for songs.

Accordingly, there is an increasing need for a method and apparatus for easily searching for a desired song without pressing a dial and inputting song titles one by one.

Korean Utility Model Registration No. 20-0202916

The present invention provides a voice recognition music selection service providing method and a voice recognition music selection apparatus that enable easier and more accurate song retrieval.

According to one aspect of the present invention, a step of providing a plurality of reference tokens generated by segmenting a reference utterance containing a song title and a singer name, segmenting on the basis of parts of speech, and stems and endings, a user a step of inputting a search utterance sentence uttered by a to search for a song in the form of text; a step of dividing the search utterance sentence into words, a part of speech, and a step of generating a plurality of search tokens by dividing the word stem and ending into a plurality of search tokens; comparing the search token with a plurality of reference tokens; calculating a similarity score of the search utterance sentence to the reference utterance sentence according to the comparison result; and providing song information to the user based on the similarity score. A method for providing a cognitive music selection service is provided.

The plurality of reference tokens are obtained by firstly dividing the reference utterance sentence based on the spaces, secondarily dividing the firstly divided reference utterance sentence based on the part of speech, and dividing the secondarily divided reference utterance sentence into verbs and adjectives. It may be generated by tertiary segmentation based on the stem and the ending.

The step of providing a plurality of reference tokens may further provide a first character of each of the plurality of reference tokens and an extension token constructed by sequentially adding one character to the first character.

The step of inputting the search utterance in the form of text may include the step of inputting the search utterance by voice, which is uttered by the user for the song search, and the step of converting the search utterance into the form of text. .

The step of inputting the search utterance in the form of text includes, after the step of converting the search utterance into the form of text, comparing the search utterance with the song title and singer name to correct errors in the search utterance. It may contain further.

In the step of generating a plurality of search tokens, the primary division of the search utterance sentence is performed on the basis of the spaces, the primary division of the search utterance sentence is secondary on the basis of the part of speech, and the second division of the search utterance sentence is performed. Multiple search tokens may be generated by tertiary segmentation of verbs and adjectives based on stems and endings.

In the step of calculating the similarity score, the similarity score may be calculated when the plurality of search tokens are all included in the plurality of reference tokens.

In the step of calculating the similarity score, a unit score may be calculated for each search token that matches the reference token, and the unit scores may be summed to calculate the similarity score.

A higher unit score may be calculated as the number of characters of the search token that matches the reference token increases.

According to one aspect of the present invention, a reference token providing unit provided with a plurality of reference tokens generated by segmenting a reference utterance sentence including a song title and a singer name on the basis of parts of speech, and stems and endings; An input part for inputting search utterance sentences uttered for song search in the form of text, a search token for generating a plurality of search tokens by dividing the search utterance sentences into parts, parts of speech, and word stems and endings. a generation unit, a token comparison unit that compares a plurality of search tokens with a plurality of reference tokens, a score calculation unit that calculates a similarity score of the search utterance sentence to the reference utterance sentence based on the comparison results, A voice recognition music selector is provided that includes an information providing results provider.

According to the present invention, it is possible to perform song searches more easily and accurately.

FIG. 1 is a diagram showing a voice recognition music selection service providing system according to an embodiment of the present invention. FIG. 2 is a flow chart showing a method of providing a voice recognition song selection service according to an embodiment of the present invention. FIG. 3 is a flowchart showing the steps of inputting a search utterance in the form of text according to an embodiment of the present invention. FIG. 4 is a diagram showing a process of determining whether or not to calculate a similarity score in the step of calculating a similarity score according to an embodiment of the present invention. FIG. 5 is a flowchart showing a similarity score calculation process in the step of calculating a similarity score according to an embodiment of the present invention. FIG. 6 is a diagram showing the configuration of a voice recognition music selection apparatus according to an embodiment of the present invention.

Since the present invention can be modified in various ways and can have various embodiments, specific embodiments will be illustrated in the drawings and described in detail in the detailed description. However, this is not intended to limit the invention to any particular embodiment, but should be understood to include all transformations, equivalents or alternatives falling within the spirit and scope of the invention. In the description of the present invention, when it is determined that a detailed description of related known technology may obscure the gist of the present invention, the detailed description will be omitted.

Although the terms first, second, etc. may be used to describe various components, the components should not be limited by the terms. The above terms are only used to distinguish one component from another.

The terminology used in the present invention is only used to describe particular embodiments and is not intended to be limiting of the invention. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this application, terms such as "including" or "having" are intended to specify the presence of the features, numbers, steps, acts, components, parts, or combinations thereof set forth in the specification. and does not preclude the presence or addition of one or more other features, figures, steps, acts, components, parts or combinations thereof.

Hereinafter, embodiments of the voice recognition music selection service providing method and the voice recognition music selection device (100) according to the present invention will be described in detail with reference to the accompanying drawings. are assigned the same drawing numbers, and redundant description thereof will be omitted.

First, a speech recognition music selection service providing system according to an embodiment of the present invention will be described.

According to this embodiment, as shown in FIG. 1, an input device for inputting search utterances uttered by a user (10) for song search, and a conversion server (server) for converting the search utterances into text form. , a generation server that divides the search utterance sentence converted into the form of text into words, the part of speech, and the word stem and ending to generate a plurality of search tokens; , and a plurality of reference tokens generated by dividing based on word stems and endings, comparing the plurality of reference tokens with the plurality of search tokens, and determining the similarity of the search utterance sentence to the reference utterance sentence according to the comparison result an analysis server that calculates a score and generates song information to be provided to the user (10) based on the similarity score; A voice recognition music selection service providing system is provided, including an accompaniment device and an input device that perform at least one of reproduction, a generation server, an analysis server, and a transmission/reception device that transmits and receives data between the accompaniment devices.

Voice data, which is a search utterance sentence uttered by the user (10) for song search, is sent by the input device to the transmission/reception device, and the corresponding voice data is sent by the transmission/reception device to the conversion server and converted into a text form. becomes text data.

Also, the corresponding text data is received by the transmitting/receiving device again and transmitted to the generation server, and at this time, is divided into a plurality of search tokens based on the punctuation, the part of speech, and the stem and ending of the word, and the plurality of search tokens are transmitted and received. sent by the device to the analysis server.

Thereby, the plurality of search tokens are divided into a plurality of reference tokens generated by dividing the reference utterance sentence including the song title and the singer name stored in the analysis server, and dividing the part of speech, and the stem and ending of the word. After the comparison, the analysis server calculates a similarity score according to the corresponding comparison result, generates song information to be provided to the user (10) based on the similarity score, and transmits the same to the transmitting/receiving device.

After that, when the user (10) selects a desired song from the song information provided to the user (10) by the transmitting/receiving device, the corresponding information is transmitted to the accompaniment device, and finally the accompaniment of the corresponding song is started. become.

According to this embodiment, it is possible to search songs more easily and accurately. Matters regarding further configurations or functions related to the above-described devices and servers depend on the method of providing the voice recognition music selection service and the content of the voice recognition music selection device 100, which will be described later.

Next, a method for providing a voice recognition music selection service according to an embodiment of the present invention will be described.

According to the present embodiment, as shown in FIG. 2, a plurality of reference tokens are provided by dividing a reference utterance sentence including a song title and a singer name, and dividing it based on parts of speech, stems, and endings. a step of inputting a search utterance sentence uttered by the user (10) for song search in the form of text (S120); dividing the search utterance sentence into parts of speech, and segmenting it based on the part of speech and the stem and ending of a word; a step of generating a plurality of search tokens (S130), a step of comparing the plurality of search tokens with a plurality of reference tokens (S140), and a step of calculating a similarity score of the search utterance sentence to the reference utterance sentence according to the comparison results (S150 ), and providing song information to the user (10) based on the similarity score (S160).

According to this embodiment, the search is performed by uttering the contents of the song to be searched, that is, by just saying the song. Multiple search tokens generated by splitting the search utterances with the same name are compared with multiple reference tokens generated by splitting the reference utterances containing song titles and singer names to derive search results. Accurate song search is possible.

Hereinafter, each step of the voice recognition music selection service providing method according to the present embodiment will be described with reference to FIGS. 2 to 6. FIG.

The step of providing a plurality of reference tokens (S110) includes providing a plurality of reference tokens generated by segmenting a reference utterance including the song title and singer name, and dividing the part of speech, and the stem and ending of the sentence. good too.

As a result, data called a plurality of reference tokens having all contents related to song information such as song titles and singer names are formed, and the corresponding data are accumulated to form one data field for comparison with the user's search request contents. (data field) can be formed.

In this method, separate data fields are provided for each content related to song information such as song titles, singer names, etc., and each data field is individually compared with the search request content, and then added up. Since the comparison with the content of the search request is performed in one data field at a time compared to the method of deriving the result, it is possible to reduce errors that may occur due to the individual comparison. Reducing the occurrence of errors allows for more accurate comparisons and derivation of results.

Here, a "token" is a single character or a group of a plurality of characters generated by dividing a spoken sentence based on the part of speech, word stem and ending, and is a data field for comparison in the present invention. can be understood to mean a unit of data that constitutes the

In addition, the ``spoken sentence'' can be understood to mean ``a sentence in which words spoken with sound are written in letters, etc.''.

For example, when the user (10) tries to search for the song "I have a lover" by the singer "Lee Eun Mi", the contents of what the user (10) says are written in characters "Lee Eun Mi has a lover." ) can be said to be an uttered sentence.

The plurality of reference tokens are obtained by firstly dividing the reference utterance sentence based on the spaces, secondarily dividing the firstly divided reference utterance sentence based on the part of speech, and dividing the secondarily divided reference utterance sentence into verbs and adjectives. It may be generated by tertiary segmentation based on the stem and the ending. More specifically, tertiary segmentation based on word stems and word endings may be performed at the character level.

As a result, a plurality of reference tokens may be provided by dividing a plurality of words constituting the reference utterance sentence based on the meaning and the like. Compared to tokens, more effective results can be derived when searching for songs.

For example, ``Look for Lee Eun-mi's lover'' is primarily divided into ``Lee Eun-mi's'', ``Lover'', ``I'm here'' and ``Look for me'' based on the phrasing, and this is also divided based on the part of speech `` Iunmi” (noun), “no” (particle), “lover” (noun), “imasu” (verb) and “look for me” (verb). "Masu" and "Search for" are based on the stem and the ending of the word, "i" (the stem), "masu" (the ending) and "search" (the stem), "shite" (the ending), "kure" (the stem) + ending). This generates the reference tokens ``iunmi'', ``no'', ``lover'', ``i'', ``masu'', ``search'', ``shite'' and ``kure''.

In the step of providing a plurality of reference tokens (S110), the first character of each of the plurality of reference tokens and an extension token configured by sequentially adding one character to the first character may be further provided. good.

As a result, in addition to the plurality of reference tokens, the data of the first character and extension token of each of the plurality of reference tokens are added to the data field, so that more detailed and accurate comparison with search tokens and song search results to be described later can be obtained. A derivation of can be performed.

For example, in addition to the reference token "iunmi", the first letter "i" and the extended tokens "iun" and "iunmi" may be provided. In this process, duplicate tokens 'Lee Eun Mi' may be selectively deleted.

The step of inputting the search utterance in the form of text (S120) is, as shown in FIGS. good too.

As a result, since the user (10) can first search for songs just by speaking, it becomes possible to search for songs more easily.

More specifically, in the step of inputting the search utterance in the form of text (S120), as shown in FIG. A step (S122) and a step (S124) of converting the retrieved utterance into text form may also be included.

In this case, the user (10) may utter a search utterance containing at least one of the song title and the singer name, preferably the search utterance containing the song title and the singer name for more accurate song search. (For example, "Find Lee Eun-mi's girlfriend.") to make a search request.

Also, the step (S124) of converting the retrieval utterance into a text form prepares source data for generating a plurality of retrieval tokens, which will be described later.

More specifically, in the step of converting the retrieved utterance into text form (S124), song information data including song titles and singer names can be used to increase accuracy in converting the retrieved utterance into text form. More specifically, the conversion accuracy can be improved by setting the above song information data as a reference value for conversion and performing conversion.

In other words, when converting a spoken sentence into text based on commonly used words, it may be converted into words other than the intended words, which may reduce the conversion accuracy. If the song information data is used as a reference for conversion, conversion accuracy can be further improved because a reference corresponding to the user's (10) utterance intention for searching for a song is provided.

In this case, artificial intelligence (AI) technology can be used, for example, by improving the sensitivity to words that are frequently exposed by deep learning, conversion accuracy can be further improved. can be done.

As shown in FIG. 3, the step of inputting the search utterance in the form of text (S120) includes the step of converting the search utterance into text form (S124), followed by the step of converting the search utterance into text form (S124). A step of comparing and correcting errors in the retrieved utterance (S126) may be further included.

In other words, even if there are some errors in the retrieval utterance sentence, the errors in the retrieval utterance sentence are corrected by comparing with the song title and singer name, and by preparing the text for generating the retrieval token, A more accurate song search becomes possible.

For example, if there is an error "is" in the search utterance sentence, such as "Lee Eun Mi's lover, please look for me", the singer name "Lee Eun Mi" and the song title "I have a lover" are compared with the search utterance sentence. By doing so, it is possible to correct the above error and prepare the text for 'Find Lee Eun Mi's lover'.

More specifically, in the step of correcting errors in the retrieved utterance (S126), a score can be calculated by comparing the retrieved utterance with the title of the song and the name of the singer. The search utterance sentence can be corrected based on the song information with the highest score among them.

As a result, the possibility of correcting the search utterance sentence to match the intention of the user (10) is further improved.

Also in this case, as described above, artificial intelligence technology, such as deep learning, can be applied, and by learning words that are frequently searched, more accurate corrections can be made to the sentences containing the relevant words. can be done.

In the step of inputting the search utterance by voice (S122), as shown in FIG. 6, the search utterance may be input by a receiver capable of speech recognition (122).

More specifically, the receiver (122) may be capable of noise filtering and/or audio amplification to increase speech recognition rate in noisy situations.

Accurate recognition of the user's (10) voice is important, as venues such as karaoke, where the present invention may be primarily practiced, are spaces with a lot of ambient noise, such as the sound of singing.

Therefore, the receiver (122) filters ambient noise and/or amplifies the user's (10) voice to improve the voice recognition rate of the user (10), thereby improving the accuracy of song search. can be made

The step of generating a plurality of search tokens (S130) may generate a plurality of search tokens by segmenting the search utterance sentence based on parts of speech, and stems and endings. Thus, by preparing multiple reference tokens and corresponding multiple search tokens, it is possible to improve effectiveness when comparing both tokens later.

In the step of generating a plurality of search tokens (S130), the search utterance sentence is primarily divided on the basis of spaces, the primarily divided search utterance sentence is secondarily divided on the basis of the part of speech, and the second-order divided search utterance sentence is A plurality of search tokens may be generated by tertiary division of verbs and adjectives based on stems and endings.

As a result, a plurality of search tokens are prepared by dividing a plurality of words constituting a search utterance sentence based on the meaning and the like. , more effective search results can be derived.

A specific example of this can refer to the example of generating a plurality of reference tokens described above.

The step of comparing (S140) with a plurality of reference tokens may compare a plurality of search tokens with a plurality of reference tokens. Accordingly, a reference data value is prepared for calculating a similarity score and providing a result thereof, which will be described later.

Also, the step of comparing multiple search tokens with multiple reference tokens (S140) may be performed via a separately provided comparison server.

The step of calculating the similarity score (S150) may calculate the similarity score of the search utterance sentence with respect to the reference utterance sentence based on the comparison results described above.

Through this similarity score, it is possible to select song information from search results and provide it to the user (10), as will be described later, so that the user (10) can obtain more effective search results. can be provided.

More specifically, in the step of calculating the similarity score (S150), as shown in FIG. 4, the similarity score may be calculated when a plurality of search tokens are all included in a plurality of reference tokens.

As a result, even if a plurality of search tokens are only partially included in a plurality of reference tokens, the calculation of the similarity score prevents unnecessary song information from being included in the search results, thereby providing more accurate song information. Search results can be provided to the user (10).

More specifically, in the step of calculating the similarity score (S150), as shown in FIG. 4, when some of the plurality of search tokens are included in the plurality of reference tokens, the user (10) is asked to repeat the search utterance sentence. You may ask to speak.

In this case, the user (10) may respeak except for the letters or words that are not expected to be contained within the multiple reference tokens, so that the multiple search tokens all fit into the multiple reference tokens. Once included, a similarity score can be calculated.

For example, when the user (10) utters ``Look for Lee Eun-mi's lover,'' the search token ``desu'' (ending) is included in multiple reference tokens for ``Lee Eun-mi's lover, please look for me.'' Therefore, the user (10) can be requested to reinstate the search utterance.

Further, in the step of calculating the similarity score (S150), as shown in FIG. 4, the user (10) is requested to utter the search utterance sentence again, and the search tokens not included in the plurality of reference tokens are You may provide information.

As a result, it is possible to prevent meaningless repetition of the search utterance sentence re-request and the user's (10) re-reproduction step, thereby improving the ease and accuracy of song retrieval.

For example, when the user (10) utters ``Lee Eun-mi's girlfriend, please find me'', the user (10) is requested to reproduce the search utterance sentence, and ``is not included in multiple reference tokens'' ” (suffix) search token information can be presented.

In the similarity score calculation step (S150), as shown in FIG. 5, a unit score is calculated for each of the search tokens that match the reference token, and the unit scores are summed to calculate the similarity score. You may

That is, a unit score may be assigned to each search token that matches the reference token, and the similarity score as the final score may be calculated by adding up all of the above unit scores.

More specifically, a higher unit score may be calculated as the number of characters of the search token that matches the reference token increases.

In other words, if a search token with a large number of characters is matched with a reference token, it means that there is a high possibility that the search token is a key word forming song information desired by the user (10). By assigning a high unit score to a token, it is possible to impart non-uniformity to the finally calculated similarity score.

For example, the search token "lover", which has more characters than the search token "i", is seen as a keyword that makes up the song title when viewed as a whole song title. By giving the unit score, it is possible to set the "lover" to contribute more to the calculation of the similarity score.

The step of providing song information (S160) may provide song information to the user (10) based on the similarity score.

As a result, the accuracy and appropriateness of providing the song search results can be improved, compared to indiscriminately providing all search results to the user (10).

More specifically, in the step of providing song information (S160), song information including song titles and singer names whose similarity score is equal to or greater than a preset value may be provided to the user (10).

That is, by providing the user (10) with song information including song titles and singer names with scores equal to or higher than the preset similarity scores, the accuracy and appropriateness of song search results are further enhanced. be able to.

In the step of providing song information (S160), song information may be provided to the user (10) in descending order of similarity scores.

Accordingly, the user (10) can quickly find the desired song by searching for song information with a high similarity score among the song information including the song title and singer name with the provided scores.

A voice recognition music selection device (100) according to an embodiment of the present invention will be described.

According to the present embodiment, as shown in FIG. 6, a criterion provided is a plurality of criterion tokens generated by segmenting a criterion utterance containing a song title and a singer name, and segmenting it based on parts of speech, and stems and endings. A token providing unit (110), an input unit (120) for inputting search utterance sentences uttered by the user (10) for song search in the form of text, dividing the search utterance sentences, parts of speech, and stems and endings as criteria. a search token generation unit (130) that generates a plurality of search tokens by dividing into a plurality of search tokens; a token comparison unit (140) that compares the plurality of search tokens with a plurality of reference tokens; Provided is a voice recognition music selection device (100) including a score calculation unit (150) for calculating a similarity score and a result providing unit (160) for providing song information to a user (10) based on the similarity score. .

According to this embodiment, the user (10) can more easily search for songs and obtain more accurate song search results.

Hereinafter, each configuration of the voice recognition music selection apparatus (100) according to the present embodiment will be described with reference to FIGS. 2 to 6. FIG.

The reference token providing unit (110) may be provided with a plurality of reference tokens generated by segmenting a reference utterance containing a song title and a singer name, and segmenting on the basis of parts of speech, and stems and endings.

In this case, a plurality of reference tokens may be provided in the form of download from a server or the like, and in a state in which the plurality of reference tokens are stored in the server or the like, a search token, which will be described later, is sent to the server. The plurality of reference tokens may be transmitted and used.

The plurality of reference tokens are obtained by firstly dividing the reference utterance sentence based on the spaces, secondarily dividing the firstly divided reference utterance sentence based on the part of speech, and dividing the secondarily divided reference utterance sentence into verbs and adjectives. It may be generated by tertiary segmentation based on the stem and the ending.

The reference token providing unit (110) may further be provided with first characters of each of the plurality of reference tokens and extended tokens formed by sequentially adding one character to the first character.

The input unit 120 may receive a search utterance sentence uttered by the user 10 for song search in the form of text.

More specifically, the input unit 120 includes a voice input unit 120 for inputting a search utterance sentence uttered by the user 10 for song search, and a search utterance sentence converted into a text form. may include a text conversion unit that

The conversion of the retrieval utterance into text form may be performed via a separate server, and for this purpose, the retrieval utterance, which is input by voice, may be transmitted to the server.

More specifically, the text conversion unit may utilize song information data including song titles and singer names in order to increase accuracy in converting the retrieved utterances into text form. By setting the information data as a reference value for conversion and performing conversion, the conversion accuracy can be improved.

In this case, artificial intelligence technology may also be used. For example, deep learning can be used to improve sensitivity to words that are frequently exposed, thereby further improving conversion accuracy.

The input unit (120) may further include an error correction unit that compares the retrieved utterance with the song title and singer name and corrects errors in the retrieved utterance.

More specifically, the error correction unit can compare the retrieved utterance text with the song title and the singer name to calculate the score, and the song information with the highest score among the song information whose comparison score exceeds a preset reference value is used as the reference. You may modify the search utterance sentence to

Also in this case, as described above, artificial intelligence technology, such as deep learning, may be applied, so that through learning of frequently searched words, sentences containing the relevant words may be corrected more accurately. can be done.

The speech input unit (120), as shown in FIG. 6, may receive a search speech sentence by a receiver (122) capable of speech recognition.

In this case, the receiver (122) may include any input device capable of inputting and recognizing voice. Said receiver (122) may include a receiver or a microphone provided at the user's (10) terminal associated with the device.

More specifically, the receiver (122) may be capable of filtering noise and/or amplifying speech to increase the recognition rate of speech in noisy situations.

The search token generation unit (130) may generate a plurality of search tokens by dividing the search utterance sentence into words, parts of speech, and stems and endings.

Alternatively, the search utterance text converted into text may be transmitted to a server provided separately, and the plurality of search tokens may be generated by the corresponding server.

A search token generation unit (130) primarily divides the search utterance sentence based on the spaces, secondarily divides the primarily divided search utterance sentence based on the part of speech, and verbs of the secondarily divided search utterance sentence. A plurality of search tokens may be generated by tertiary segmentation of adjectives and adjectives based on their stems and endings.

The token comparator (140) may compare multiple search tokens with multiple reference tokens.

In this case, multiple search tokens may be cross-compared with multiple reference tokens in a separate server, and multiple search tokens may be sent to said server for this purpose.

The score calculation unit (150) may calculate a similarity score of the search utterance sentence to the reference utterance sentence based on the comparison result described above.

More specifically, the score calculation unit (150) may calculate the similarity score when all of the plurality of search tokens are included in the plurality of reference tokens, as shown in FIG.

As shown in FIG. 4, when some of the plurality of search tokens are included in the plurality of reference tokens, the score calculation unit (150) requests the user (10) to utter the search utterance sentence again. good.

Also, as shown in FIG. 4, the score calculation unit 150 requests the user 10 to utter the search utterance sentence again and provides information on search tokens not included in the plurality of reference tokens. may

As shown in FIG. 5, the score calculation unit (150) may calculate a unit score for each search token that matches the reference token, and add up each unit score to calculate a similarity score. .

More specifically, the unit score may be calculated to be higher as the number of characters of the search token that matches the reference token is larger.

The result providing unit (160) may provide song information to the user (10) based on the similarity score.

More specifically, the result providing unit (160) may provide the user (10) with song information including song titles and singer names whose similarity score is greater than or equal to a preset value.

Also, the result providing unit (160) may provide the user (10) with song information including song titles and singer names in descending order of similarity scores.

The calculation of the corresponding similarity score described above may be performed by a separate server, requesting re-request of the search utterance sentence, providing information on search tokens not included in a plurality of reference tokens, and song information as search results. A display may be provided for the provision of.

On the other hand, the constituent elements of the above-described embodiment can be easily grasped from a process point of view. That is, each component can be grasped by each process. Also, the processes of the above-described embodiments can be easily understood in terms of the components of the apparatus.

In addition, the technical contents described above may be embodied in the form of program instructions that can be executed by various computer means and recorded on a computer-readable medium. The computer-readable media may include program instructions, data files, data structures, etc. singly or in combination. The program instructions recorded on the medium may be those specifically designed and constructed for the embodiment, or they may be of the kind known and available to those of skill in the computer software arts. Examples of computer-readable recording media include magnetic media such as hard disks, proppy discs and magnetic tapes, optical media such as CD-ROMs and DVDs, and floptical disks. ), and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code, such as produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. A hardware device may be configured to act as one or more software modules to perform the operations of an embodiment, and vice versa.
An embodiment of the present invention has been described above. It can be said that the present invention can be variously modified and changed by deletion or addition, etc., and this is also included in the scope of the present invention.

10: User 100: Speech recognition music selection device 110: Reference token provider 120: Input unit 122: Receiver 130: Search token generator 140: Token comparator 150: Score calculator 160: Result provider

Claims (10)

providing a plurality of reference tokens generated by segmenting a reference utterance containing a song title and a singer name, and segmenting on the basis of parts of speech, and stems and endings;
Inputting a search utterance sentence uttered by a user for searching for a song in the form of text;
generating a plurality of search tokens by segmenting the search utterance on the basis of phrasing, parts of speech, and stems and endings;
comparing said plurality of search tokens with said plurality of reference tokens;
A method for providing a voice recognition music selection service, comprising: calculating a similarity score of the retrieved utterance sentence to the reference utterance sentence according to a comparison result; and providing song information to the user based on the similarity score. The plurality of reference tokens are
Primarily dividing the reference utterance sentence based on the spaces,
secondarily dividing the primarily divided reference utterance sentence based on the part of speech;
2. The method of claim 1, wherein the verbs and adjectives of the secondary-divided reference utterance sentence are tertiary-divided based on the stem and ending. In providing the plurality of reference tokens,
2. The provision of the speech recognition music selection service according to claim 1, further comprising a first character of each of said plurality of reference tokens and an extension token configured by sequentially adding one character to said first character. Method. The step of inputting the search utterance in the form of text includes:
2. The voice recognition music selection service according to claim 1, comprising the steps of inputting the search utterance sentence uttered by the user for the song search by voice; and converting the search utterance sentence into the text form. How to provide. The step of inputting the search utterance in the form of text includes:
After the step of converting the retrieved utterance into text form,
5. The method of claim 4, further comprising comparing the retrieval utterance with the song title and singer name to correct errors in the retrieval utterance. In generating the plurality of search tokens,
Primarily dividing the retrieved utterance sentence based on the spaces between words;
secondarily dividing the primarily divided retrieval utterance sentence based on the part of speech;
2. The method of claim 1, wherein the plurality of search tokens are generated by tertiary division of verbs and adjectives in the secondary division of the search utterance sentence based on the stem and the ending. . In the step of calculating the similarity score,
2. The method of claim 1, wherein the similarity score is calculated when all of the plurality of search tokens are included in the plurality of reference tokens. In the step of calculating the similarity score,
2. The method of claim 1, wherein a unit score is calculated for each of the search tokens matched with the reference token, and the similarity score is calculated by summing the unit scores. 9. The method of claim 8, wherein the greater the number of characters of the search token matching the reference token, the higher the unit score calculated. a reference token provider provided with a plurality of reference tokens generated by segmenting a reference utterance containing a song title and a singer name, and segmenting on the basis of parts of speech, and stems and endings;
an input unit for inputting a search utterance sentence uttered by a user for song search in the form of text;
a search token generation unit that generates a plurality of search tokens by dividing the search utterance sentence based on the phrasing, part of speech, and stem and ending;
a token comparator that compares the plurality of search tokens with the plurality of reference tokens;
A voice recognition music selection device, comprising: a score calculation unit that calculates a similarity score of the retrieved utterance sentence to the reference utterance sentence based on a comparison result; and a result providing unit that provides song information to the user based on the similarity score.

Images