JP2023006055A - Program, information processing device, and method - Google Patents

Abstract

To make it possible to estimate reading and accent according to context.SOLUTION: There is provided a program for operating a computer (10) provided with a processor (11), which causes the processor to execute the steps of: acquiring learning data including language data and speech language data representing the language data by a speech language defined to simultaneously represent reading and accent (S101); learning a translation model that outputs the speech language data when the language data is input using the learning data (S102); and outputting the learned translation model (S103).SELECTED DRAWING: Figure 5

Description

The present disclosure relates to programs, information processing apparatuses, and methods.

Conventionally, techniques for synthesizing speech from language data have been developed. In Patent Document 1, "The acoustic feature amount estimation unit 42 of the speech synthesizer 1 prepares sentences representing the contents of an utterance by extracting characters or character strings representing how to read the contents of the utterance, prosody symbols representing the prosody, and features given to the utterance. The text data described by a character string using one or both of the utterance style symbols representing the estimates the speech waveform using the estimated acoustic feature quantity.The acoustic feature quantity generation model has an encoder and decoder using DNN.The encoder uses the RNN to match the utterance content indicated by the text data in the sentence. A character string feature amount is generated in consideration of character strings before and after the content of the utterance, and the decoder generates a sound corresponding to the utterance content based on the feature amount generated by the encoder and the sound feature amount generated in the past by the RNN. Generating a feature amount” technology is disclosed.

There is also a technique for estimating the pronunciation and accent of an input text in order to synthesize speech from the text. For example, for the text "Malaysia no Mizu", there is a model that estimates the reading "Mareishi Amizu" and a model that estimates the accent "12211112" (accented expression) (https://sites.google.com/site /suzukimasayuki/accent).

Japanese Patent Application Laid-Open No. 2020-034883

However, in the prior art, reading and accent can be estimated separately, but this is done by estimating words and readings by performing morphological analysis on the text and applying known accents of the words. ing. Therefore, there is a problem that it is difficult to estimate the accent according to the context. Moreover, there is a problem that the pronunciation and accent cannot be estimated for new words for which correct accent data do not yet exist.

The purpose of this disclosure is to enable contextual reading and accent estimation

Therefore, a technology is provided that can estimate the reading and accent according to the context.

A program according to the present disclosure is a program for operating a computer having a processor, wherein the processor is provided with utterances expressing the linguistic data in an utterance language defined so as to express linguistic data and pronunciation and accent at the same time. acquiring learning data including language data; using the learning data to learn a translation model that outputs the spoken language data when the language data is input; and outputting the learned translation model. and let it run.

According to the present disclosure, contextual reading and accent can be estimated.

1 is a block diagram showing the configuration of an information processing system 1; FIG. 2 is a block diagram showing the functional configuration of the information processing device 10; FIG. It is a figure which shows the structural example of a translation model. 4 is a diagram showing an example of a screen displayed on the user terminal 20; FIG. 4 is a flowchart showing an example of the flow of learning processing by the information processing apparatus 10; 4 is a flow chart showing an example of the flow of performing voice synthesis processing by the information processing device 10. FIG. It is a figure which shows the example of an utterance language expression.

Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. In the following description, the same parts are given the same reference numerals. Their names and functions are also the same. Therefore, a detailed description thereof will not be repeated.

<Summary of this disclosure>
The present disclosure describes a program, an information processing device, and a method for synthesizing speech from text data input by a user and reproducing synthesized speech. The present disclosure also describes a program, etc., for learning a translation model for estimating spoken language data according to the present disclosure from text data.

<1. Configuration of Information Processing System 1>
An information processing system 1 according to the present disclosure will be described with reference to FIG. The information processing system 1 according to the present disclosure synthesizes speech for text data input by a user, and reproduces synthesized speech.

FIG. 1 is a diagram showing the configuration of an information processing system 1. As shown in FIG. The information processing system 1 includes an information processing device 10 , a user terminal 20 and a network 30 .

The information processing device 10 according to the present disclosure is a device for executing learning processing for learning a translation model, speech synthesis processing for synthesizing speech, and the like. The information processing apparatus 10 is, for example, a laptop computer, a rack-mount type computer, a tower type computer, or the like. The information processing device 10 may be configured by a plurality of information processing devices 10 and the like. The method of distributing a plurality of functions required for realizing the information processing system 1 can be appropriately determined in consideration of the processing capability of each piece of hardware, the specifications required for the information processing system 1, and the like.

The information processing device 10 includes a processor 11 , a memory 12 , a storage 13 , a communication IF 14 and an input/output IF 15 .

The processor 11 is hardware for executing an instruction set described in a program, and is composed of an arithmetic unit, registers, peripheral circuits, and the like.

The memory 12 temporarily stores programs and data processed by the programs, and is a volatile memory such as a DRAM (Dynamic Random Access Memory).

The storage 13 is a storage device for storing data, and is, for example, a flash memory, HDD (Hard Disc Drive), or SSD (Solid State Drive).

The communication IF 14 is an interface for inputting/outputting signals so that the information processing device 10 communicates with an external device. The communication IF 14 is wired or wirelessly connected to a network 30 such as the Internet or wide area Ethernet.

The input/output IF 15 functions as an interface with an input device (for example, a pointing device such as a mouse, a keyboard) for receiving input operations, and an output device (display, speaker, etc.) for presenting information.

The user terminal 20 is, for example, a computer such as a laptop computer, a smart phone, or a tablet.

The information processing device 10 and the user terminal 20 are configured to be able to communicate with each other via the network 30 .

<1.2. Configuration of Information Processing Device 10>
FIG. 2 is a block diagram showing the functional configuration of the information processing device 10. As shown in FIG. As shown in FIG. 3, the information processing apparatus 10 includes a communication section 110, a storage section 120, and a control section .

The communication unit 110 performs processing for the information processing device 10 to communicate with an external device.

The storage unit 120 stores data and programs used by the information processing apparatus 10 . The storage unit 120 stores a learning data DB 121, a model DB 122, and the like.

The learning data DB 121 is a database that holds learning data. The learning data includes linguistic data and spoken language data expressing the linguistic data in a spoken language defined so as to simultaneously represent pronunciation and accent. The linguistic data is linguistic data to be subjected to speech synthesis, such as text data and voice data. Details of the learning data will be described later.

The model DB 122 is a database that holds translation models and translation model parameters. The parameters of the translation model held by the model DB 122 are updated each time the translation model is learned by the learning unit 133, which will be described later. In addition, the model DB 122 holds initial parameters before the translation model is learned by the learning unit 133 .

The control unit 130 performs functions shown in the reception control unit 131, the transmission control unit 132, the learning unit 133, the input unit 134, the translation unit 135, and the synthesis unit 136 by the processor 11 of the information processing device 10 performing processing according to the program. demonstrate.

The reception control unit 131 controls processing for the information processing device 10 to receive a signal from an external device according to a communication protocol.

The transmission control unit 132 controls processing for transmitting a signal from the information processing device 10 to an external device according to a communication protocol.

The learning unit 133 uses learning data to learn a translation model that outputs spoken language data when language data is input.

Specifically, the learning unit 133 first acquires learning data from the learning data DB 121 . The learning data includes linguistic data and spoken language data expressing the linguistic data in a spoken language defined so as to simultaneously represent pronunciation and accent.

The linguistic data is linguistic data to be subjected to speech synthesis, such as text data and voice data. In the present disclosure, a case where language data is text data will be described as an example. If the language data is voice data, the information processing apparatus 10 may be configured to convert the voice data into text data by voice analysis.

Spoken language is defined to represent reading and accent at the same time. Conventionally, the pronunciation and accent before extracting the speech feature amount are expressed by combining the language data expressed only by the pronunciation and the language data expressed only by the accent. For example, conventionally, the text "Malaysian Water" was represented by the reading "Malaysia Amizu" and the accent "12211112". A 1 in this accent corresponds to a lowered note and a 2 to a raised note. However, this makes it difficult to estimate the accent along the context. In addition, the separation of pronunciation and accent reduces the learning efficiency of the translation model. Therefore, the spoken language of the present disclosure is newly defined as a representation of the pronunciation and the accent, without separately including the representation of the linguistic data only by the pronunciation and the representation of the linguistic data only by the accent.

Specifically, the spoken language of the present disclosure is defined such that, for one sound of language data, the reading of the one sound and the accent of the one sound are represented at once by the same symbol. The defined spoken language has the following rules.
・Hiragana "-" corresponds to the accent "1 (down)".
・The katakana "~" corresponds to the accent "2 (up)".

For example, for the above text "Malaysian water", when there is a reading (mareishi anmizu) and an accent (12211112),
・The accent corresponding to "ma" is "1 (down)", so the hiragana "ma"
・The accent corresponding to "re" is "2 (up)", so the katakana "re"
・The accent corresponding to "-" is "2 (up)", so the wavy line "-"
・The accent corresponding to "shi" is "1 (down)", so the hiragana "shi"
・The accent corresponding to "a" is "1 (down)", so the hiragana "a"
・The accent corresponding to "no" is "1 (down)", so the hiragana "no"
・The accent corresponding to "mi" is "1 (rising)", so the hiragana "mi"
・The accent corresponding to "zu" is "2 (go up)", so the hiragana "zu"
becomes. Therefore, in the said spoken language, "Malaysian water" becomes "Male-Shiamizu". In this way, spoken languages use two different ways of expressing each sound depending on the accent. In addition, this is not limited to Japanese, and in other languages, for example, depending on the accent, the international phonetic symbols that represent the reading can be used properly by adding accent marks or inverting them. Just do it.

Spoken language may also be defined to include emotional expressions and prosodic support symbols. Emotional expressions may be expressed, for example, by Unicode-defined pictograms (including non-faces). In addition, the prosody supporting symbols are arbitrary, such as "!" for exclamation mark, "?" for question mark, "↑" for "raise", "↓" for "lower", "←" for "push left", and so on. It can be expressed by the symbol of By including emotional expressions and prosody-supporting symbols, the accent estimation by the translation model is in line with the context, and emotional expressions are also possible.

The translation model uses learning data to output spoken language data when language data is input. FIG. 3 is a diagram showing a configuration example of a translation model. The translation model includes a first split-joint part 151 , a translation part 152 and a second split-joint part 153 .

The first dividing/combining unit 151 divides the language data into token strings. Specifically, first, when text data is input, the first splitting/joining unit 151 generates a token list for the text data using a word non-corresponding dictionary prepared in advance. A non-word correspondence dictionary is a dictionary in which words and readings do not necessarily correspond. In conventional dictionaries, reading and accent information are assigned to each word. On the other hand, in the word-uncorresponding dictionary of the present disclosure, readings of words are not necessarily registered. In addition, the word-uncorresponding dictionary of the present disclosure does not always have words corresponding to readings.

Specifically, the non-word correspondence dictionary consists of a word list, which is a list of words in natural language, and a spoken word list, which is a list of words in spoken language. A natural language word list is a list of words related to sentences containing kanji and kana, and some of the words are not associated with information on readings and accents. For example, the natural language word list includes general words such as "newspaper", "company", and "television".

The spoken word list of the spoken language is a list of spoken words whose words are character strings obtained by dividing sentences of the spoken language into appropriate lengths. The utterance word list contains words that frequently appear in utterances, and is independent of natural language words. For example, the utterance word list registers utterance words that are different from general words, such as "Katta?" Here, (emoticon) is a symbol to be handled as an emoticon in the spoken language. By introducing this word-uncorresponding dictionary, it is possible to register natural phrases and emotional expressions in an appropriately organized form. By using this uttered word list, it is possible to estimate the accent according to the context in the translation model. The utterance word list is created by manually registering utterance words with high frequency, by registering utterance words automatically optimized by a likelihood maximization algorithm, etc., or by using both. is generated by

Tokens are the smallest elements that make up a sentence or word. The first splitting/joining unit 151 splits the text data into token strings by selecting an appropriate token string from the token list. Then, first splitting/coupling section 151 outputs the token string to translation section 152 .

Also, when a token string is input from the translation unit 152, the first splitting/joining unit 151 outputs text data in which the token string is joined.

When the token string of text data is input from the first splitting/joining unit 151, the translation unit 152 translates it into a token string of the spoken language. The translation unit 152 is, for example, an RNN (Recurrent Neural Network), an LSTM (Long Short Term Memory), or an Encoder-Decoder model that uses only an Attention mechanism. As an encoder-decoder model using only the attention mechanism, for example, a model called Transformer is disclosed (https://arxiv.org/pdf/1706.03762.pdf). In the present disclosure, a case where Transformer is adopted will be described as an example. Encoder-Decoder models like Transformer do not use structures such as RNN and CNN. Arbitrary layers can be used for the Encoder layer and the Decoder layer. Translating unit 152 outputs the token string of the spoken language to second splitting/coupling unit 153 .

The second dividing/combining unit 153 outputs spoken language data obtained by combining the token strings of the spoken language input from the translation unit 152 .

Also, the second splitting/joining unit 153 splits the spoken language data into token strings. Specifically, when the spoken language data is input, the second splitting/coupling unit 153 first generates a token list for the spoken language data using a natural language dictionary. The generation of this token list is performed when the translation model is learned, and is performed independently of the generation of the token list of the first splitter/joiner 151 . The second splitting/joining unit 153 splits the text data into token strings by selecting an appropriate token string from the token list. Then, second splitting and joining section 153 outputs the token string to translation section 152 .

The learning unit 133 inputs the text data to the first splitting/joining unit 151 and inputs the spoken language data corresponding to the text data to the second splitting/joining unit 153 . The learning unit 133 acquires the token string of spoken language data obtained by translating the token string of the text data by the translation unit 152 . Also, the learning unit 133 acquires the token string of the spoken language data divided by the second splitting/combining unit 153 . The learning unit 133 learns the parameters of the translation model so that the token string output from the translation unit 152 and the token string output from the second splitting/joining unit 153 match each other. Since the translation model uses an attention mechanism, it can automatically learn where to focus when expressing spoken language in the context of a natural language. As a learning method, a method similar to Transformer or any other method can be used. For example, the learning unit 133 learns the parameters of the translation model by performing this series of processes on each piece of learning data.

Then, the learning unit 133 stores the parameters of the learned translation model in the model DB 122 . In this way, the learning unit 133 learns that the natural language word list does not correspond to the spoken language list that expresses the language data in the spoken language defined so as to express the pronunciation and the accent at the same time. By using the correspondence dictionary and the spoken language data for learning the translation model, the learning efficiency can be improved by reducing the number of characters, and the reading and accent according to the context can be estimated.

The input unit 134 receives input of language data. Specifically, when the reception control unit 131 receives the text data from the user terminal 20, the input unit 134 accepts input of the text data.

FIG. 4 is a diagram showing an example of a screen displayed on the user terminal 20. As shown in FIG. As shown in FIG. 4, screen 160 includes text box 161 and button 162 .

A text box 161 is a text box for entering text data.

The button 162 is a button for transmitting text data input to the text box 161 to the information processing apparatus 10, receiving synthetic speech from the information processing apparatus 10, and reproducing the synthetic speech.

In this way, the input unit 134 of the present disclosure receives input of text data that the user wants to synthesize into speech. In addition, it is good also as a structure which does not go through a network.

The translation unit 135 uses a translation model learned in advance to translate the language data into spoken language data expressing the language data in a spoken language defined so as to simultaneously express the pronunciation and accent of the language data.

Specifically, the translation unit 135 acquires a translation model and learned parameters from the model DB 122 . Next, the translation unit 135 obtains spoken language data by inputting the text data into the translation model. Then, translation section 135 outputs the spoken language data to synthesis section 136 .

The synthesizing unit 136 extracts the speech feature amount of the language data based on the spoken language data, and performs speech synthesis based on the speech feature amount.

Specifically, the synthesizing unit 136 first acquires speech feature values by inputting the spoken language data into an arbitrary speech feature value estimation model. The speech feature estimation model is a model that outputs speech features when reading and accent are input. The speech feature quantity estimation model is, for example, DNN. The speech feature amount is, for example, a mel spectrogram. The synthesizing unit 136 may be configured to extract the pronunciation and accent of the spoken language data according to the input format of the speech feature quantity estimation model to be used.

Next, the synthesizing unit 136 synthesizes speech from the speech feature using an arbitrary vocoder. A vocoder generates a voice waveform from a voice feature amount. the vocoder. The voice waveform may be learned in advance so as to reproduce a predetermined person, character, animal, or the like. Then, the synthesizing unit 136 causes the transmission control unit 132 to transmit the synthetic voice obtained by synthesizing the voices to the user terminal 20 .

<2. Operation>
Processing in the information processing system 1 will be described below with reference to the drawings.

<2.1. Learning processing>
FIG. 5 is a flowchart showing an example of the flow of learning processing by the information processing apparatus 10. As shown in FIG. The information processing apparatus 10 executes the process at arbitrary timing. The arbitrary timing is, for example, the timing at which the operator of the information processing apparatus 10 receives a learning start signal.

In step S101, the learning unit 133 acquires from the learning data DB 121 learning data including language data and spoken language data expressing the language data in a spoken language defined so as to simultaneously express reading and accent.

In step S102, the learning unit 133 learns a translation model using learning data.

In step S103, the learning unit 133 stores the learned parameters of the translation model in the model DB 122, and ends the process. According to the learning process, the information processing apparatus 10 uses spoken language data expressing language data in a spoken language defined so as to express both pronunciation and accent at the same time for learning a translation model. As a result, the information processing apparatus 10 can learn a translation model that can estimate the reading and accent according to the context while improving the learning efficiency by reducing the number of characters.

<2.2. Speech Synthesis Processing>
FIG. 6 is a flow chart showing an example of the flow of voice synthesis processing by the information processing apparatus 10. As shown in FIG. The information processing apparatus 10 executes the process at arbitrary timing. The arbitrary timing is, for example, the timing of receiving text data from the user terminal 20, or the like.

In step S111, the reception control unit 131 receives the text data from the user terminal 20, and the input unit 134 accepts input of the text data.

In step S<b>112 , the translation unit 135 acquires a translation model and learned parameters from the model DB 122 .

In step S113, the translation unit 135 obtains spoken language data by inputting the text data into the translation model.

In step S114, the synthesizing unit 136 acquires speech features by inputting the spoken language data to an arbitrary speech feature estimation model.

In step S115, the synthesizing unit 136 synthesizes speech from the speech feature using an arbitrary vocoder.

In step S116, the synthesizing unit 136 causes the transmission control unit 132 to transmit the synthetic voice obtained by synthesizing the voices to the user terminal 20, and ends the process. In this way, the information processing apparatus 10 converts the language data into spoken language data that expresses the language data in a spoken language that is defined to simultaneously represent the reading and accent of the language data using a translation model that has been learned in advance. translate. Thereby, the information processing apparatus 10 can estimate the reading and accent according to the context.

<3. Summary>
Conventionally, a dictionary of words whose pronunciation is already known is used to estimate the pronunciation and accent. Accent estimation using such dictionaries has the problem that it cannot estimate accents for new words that do not exist, is vulnerable to accent changes according to context, and cannot handle natural accents that include emotional expressions. there were. Therefore, conventional pronunciation/accent estimation cannot achieve natural speech synthesis.

As described above, according to the present disclosure, learning data including linguistic data and spoken language data expressing the linguistic data in a spoken language defined so as to express reading and accent at the same time is acquired, and the learning data is used to learn a translation model that outputs spoken language data when language data is input, and outputs the learned translation model. This makes it possible to learn a translation model capable of estimating reading and accent in context.

In addition, according to the present disclosure, the language data is translated into spoken language data expressing the language data in a spoken language defined to simultaneously represent the reading and accent of the language data using a pre-learned translation model. . This makes it possible to estimate the reading and accent according to the context.

Further, based on the uttered language data, the speech feature amount of the language data is extracted, and based on the speech feature amount, speech synthesis is performed to obtain synthesized speech. This makes it possible to perform smooth speech synthesis in line with the context.

<Other Modifications>
Although the disclosed embodiments have been described above, they can be implemented in various other forms, and can be implemented with various omissions, substitutions, and modifications. These embodiments, modifications, omissions, substitutions and changes are included in the technical scope of the claims and their equivalents.

For example, each function of the information processing device 10 may be configured in another device. For example, each DB of the storage unit 120 may be constructed as an external database.

Also, in the above disclosure, the spoken language is defined by reading and accent, but it is not limited to this. For example, spoken language can be used to express readings, dynamics, emotions, and the like. FIG. 7 is an example of spoken language representation. For example, the spoken language can be expressed in three levels from 0 to 2 in terms of strength. In this case, not only hiragana and katakana, but also phonetic equivalents corresponding to "2" on the right side of FIG. 7 can be used to express spoken language.

<Appendix>
The items described in each of the above embodiments will be added below.
(Appendix 1) A program for operating a computer (10) comprising a processor (11), the program instructing the processor according to a speech language defined to simultaneously represent language data, reading and accent A step of acquiring learning data including spoken language data expressing the language data (S101); and a step of learning a translation model that outputs the spoken language data when language data is input using the learning data (S101). S102) and a step of outputting the learned translation model (S103).

(Supplementary Note 2) The program according to (Supplementary Note 1), wherein the spoken language does not separately include the language data expressed only by pronunciation and the language data expressed only by accents.

(Supplementary note 3) The program according to (Supplementary note 1) or (Supplementary note 2), wherein the spoken language expresses the reading of the one sound and the accent of the one sound with the same symbol for one sound of the language data. .

(Appendix 4) The program according to (Appendix 2) or (Appendix 3), wherein the translation model is an Encoder-Decoder model using only an Attention mechanism.

(Appendix 5) In the learning step, the translation model is generated using a first word list that is a list of natural language words, a second word list that is a list of spoken language words, and the learning data. (Appendix 1)-( A program according to any one of Appendix 4).

(Additional remark 6) A program for operating a computer (10) having a processor (11), the program comprising a step of accepting input of language data (S111), and pre-loading the language data to the processor Using the learned translation model, translating the language data into spoken language data expressing the spoken language in a spoken language defined to simultaneously express the pronunciation and accent of the language data (S112); and outputting the spoken language data. and wherein said translation model, upon receiving language data, outputs said spoken language data.

(Supplementary Note 7) A step of extracting speech feature amounts of the language data based on the spoken language data (S114), and a step of obtaining synthesized speech by performing speech synthesis based on the speech feature amounts (S115 ), and outputting the synthesized speech in the step of outputting (S116).

(Appendix 8) An information processing device (10) having a processor (11), comprising language data and spoken language data expressing the language data in a spoken language defined to simultaneously express reading and accent. a step of acquiring learning data (S101); a step of learning a translation model for outputting said spoken language data when language data is input using said learning data (S102); and a step of outputting said learned translation model. An information processing device that executes step (S103).

(Appendix 9) A computer (for example, the information processing device 10) acquires learning data including language data and spoken language data expressing the language data in a spoken language defined so as to express both pronunciation and accent at the same time. a step (S101), a step (S102) of learning a translation model that outputs the spoken language data when language data is input using the learning data, and a step (S103) of outputting the learned translation model; , how to run.

1 information processing system, 10 information processing device, 11 processor, 12 memory, 13 storage, 14 communication IF, 15 input/output IF, 20 user terminal, 30 network, 110 communication unit, 120 storage unit, 121 learning data DB, 122 model DB, 130 control unit, 131 reception control unit, 132 transmission control unit, 133 learning unit, 134 input unit, 135 translation unit, 136 synthesis unit.

Claims (9)

A program for operating a computer comprising a processor, the program causing the processor to:
acquiring learning data including language data and spoken language data representing the language data in a spoken language defined to simultaneously represent pronunciation and accent;
using the learning data to learn a translation model that outputs the spoken language data when language data is input;
a step of outputting the learned translation model;
program to run. The spoken language does not separately include a representation of the linguistic data in pronunciation only and a representation of the linguistic data in accent only,
A program according to claim 1. In the spoken language, for one sound of the language data, the reading of the one sound and the accent of the one sound are represented by the same symbol,
3. A program according to claim 1 or 2. 4. The program according to claim 2, wherein said translation model is an Encoder-Decoder model using only an Attention mechanism. in the learning step, learning the translation model using a first word list that is a list of natural language words, a second word list that is a list of spoken language words, and the learning data;
wherein the first word list is a reading and unaccented list of natural language words;
the second word list is a list of words in a spoken language;
A program according to any one of claims 1 to 4. A program for operating a computer comprising a processor, the program causing the processor to:
accepting input of language data;
a step of translating the language data into spoken language data expressing the language data in a spoken language defined so as to simultaneously represent the pronunciation and accent of the language data using a pre-learned translation model;
outputting said spoken language data;
and
When the translation model receives language data, it outputs the spoken language data.
program. extracting speech features of the language data based on the spoken language data;
obtaining synthesized speech by performing speech synthesis based on the speech feature;
and
outputting the synthesized speech in the outputting step;
7. A program according to claim 6. An information processing device comprising a processor,
acquiring learning data including language data and spoken language data representing the language data in a spoken language defined to simultaneously represent pronunciation and accent;
using the learning data to learn a translation model that outputs the spoken language data when language data is input;
a step of outputting the learned translation model;
Information processing device that executes the computer
acquiring learning data including language data and spoken language data representing the language data in a spoken language defined to simultaneously represent pronunciation and accent;
using the learning data to learn a translation model that outputs the spoken language data when language data is input;
a step of outputting the learned translation model;
how to run.

Images