JP4344195B2 - Speech synthesis list generation apparatus and speech synthesis list generation program - Google Patents

Description

  The present invention relates to a speech synthesis list generation device and a speech synthesis list generation program for generating a speech synthesis list used when speech synthesis is performed.

Conventionally, as a speech synthesis method, a speech synthesis method in which phonemes and words are selected and connected in a variable-length synthesis unit from a large-scale speech database has been proposed (see, for example, Non-Patent Document 1).
In these speech synthesis methods, a list of phonemes or words that are stored in the speech database and serve as a search unit for speech synthesis is generated by, for example, the following hash method (see Non-Patent Document 2, for example). .

  For example, in order to generate a phoneme string list, this hash method sequentially reads phoneme texts from a plurality of phoneme texts in which speech is written in phoneme text data, and if the phoneme text is an existing phoneme string, The count value of the number (appearance frequency) is increased, and if it is a new phoneme string, a memory area for counting the number of phoneme strings is secured. By repeating the above operation for the phoneme text, the number (appearance frequency) of each phoneme string can be obtained, and a phoneme string list in which the phoneme strings are associated with the appearance frequencies can be generated.

Hiroyuki Seki, Kunifumi Oide, Toru Toki, “Examination of Waveform-Connected Speech Synthesis Using Words as a Search Unit”, Proc. Of the Acoustical Society of Japan, I-1-10-7, pp. 237-238, Sept. 2002 Carnihan, Richie, “Program Language C”, Kyoritsu Publishing, pp. 174-176

  However, in order to generate a phoneme string list by the above-described hash method, it is necessary to secure a memory area for counting the number of appearance frequencies for all phoneme strings written as phoneme text. is there. For this reason, if a phoneme string list is generated from a large amount of phoneme text in order to construct a large-scale speech database, there is a problem that the amount of memory used becomes large.

  Therefore, in order to actually generate a phoneme string list, the memory amount is reduced by taking measures such as not counting the number of phoneme strings longer than a certain value (usually about four). ing. However, in this case, since the phoneme string longer than a certain length is not counted, the number of phonemes having a value equal to or higher than a certain appearance frequency (threshold value) is counted for all phoneme strings existing in the phoneme text. I can't. For this reason, in reality, even a phoneme string having a high appearance frequency may not be a candidate for the phoneme string list due to the long length of the phoneme string. As a result, when attempting to search for a phoneme string having a high frequency of appearance in speech synthesis, a search is made with a short phoneme string, increasing the number of searches, and speech synthesis takes time.

  The present invention has been made in view of the above problems, and suppresses the amount of memory used when generating a phoneme string list or word string list from phoneme text or word text, and a certain appearance frequency (threshold value). (2) To provide a speech synthesis list generation apparatus and a speech synthesis list generation program capable of generating a phoneme string list or word string list obtained by extracting all phoneme strings or word strings having the above values. With the goal.

The present invention was devised to achieve the above object, and first, the speech synthesis list generating apparatus according to claim 1 is configured to generate speech from a plurality of phoneme texts in which speech is written into phoneme text data. A speech synthesis list generation device for generating a phoneme string list as a synthesis search unit, comprising: a fixed-length phoneme string extraction means; an appearance frequency accumulation means; an accumulation means; and a list output means; cumulative means cumulative meter determining means, characterized in that said list output means comprises a data deleting unit.

  According to such a configuration, the speech synthesis list generating device uses the fixed-length phoneme string extraction unit to convert phonemes corresponding to the continuous phoneme number of each phoneme text from each phoneme text so as to have a predetermined phoneme number. Extract as a column. Here, the phoneme text is a voice written into phoneme text data. For example, when the voice “news” is written in the phoneme which is the minimum unit of the speech language, the phoneme text is “ ny / u: / s / u / d / e / s / u "(" / "is a phoneme break). If, for example, “3” is set as the predetermined number of phonemes, the fixed-length phoneme string extraction unit performs continuous phonemes with the number of phonemes “3” (for example, “ny / u: / s”). And a phoneme string is extracted. The fixed-length phoneme string extraction means sequentially extracts phoneme strings having different phoneme numbers by sequentially increasing (+1) the number of phonemes.

Further, the speech synthesis list generation device uses the cumulative determination unit to generate the appearance frequency of phoneme sequences corresponding to one phoneme number shorter than the phoneme number of the phoneme sequence from the head of the phoneme sequence extracted by the fixed-length phoneme sequence extraction unit. subject but in advance when more than a threshold value set, the phoneme string extracted at fixed long sound Motoretsu extracting means determines that the object to be cumulative occurrence frequency, the frequency accumulation means, for accumulating the frequency of occurrence The frequency of appearance of phoneme strings determined to be is accumulated. This is because, when the frequency of appearance of phoneme strings shorter than the phoneme string extracted by the fixed-length phoneme string extraction means is less than a predetermined threshold value, even if the phoneme strings including the phoneme string are accumulated, the threshold value is exceeded. Because there is nothing.

For example, when accumulating the appearance frequencies of “ny / u: / s / u”, which is a phoneme string having a phoneme number “4”, “ny / u” being a phoneme string having a phoneme number “3” shorter than the phoneme string. When the appearance frequency of “u: / s” does not reach the threshold value, the appearance frequency of “ny / u: / s / u” does not exceed the threshold value. Therefore, “ny / u: / s / u” is omitted from the target of accumulating the appearance frequency.
Then, the list generation device for speech synthesis refers to the storage unit that stores the phoneme sequences for each number of phonemes by the list output unit, and outputs a phoneme sequence having an appearance frequency higher than a predetermined threshold as a phoneme sequence list. .

  The fixed-length phoneme string extraction means sequentially increases the number of phonemes to extract a phoneme string from the phoneme text, and the list output means outputs a phoneme string list having different phoneme numbers. As a result, the speech synthesis list generation device sequentially accumulates the appearance frequencies in order from the phoneme sequence having the smallest number of phonemes, and sequentially selects only the phoneme sequences including the phoneme sequences having the high appearance frequency (greater than the threshold value). Since the phoneme strings are accumulated, a phoneme string list can be generated at high speed.

The speech synthesis list generation device outputs a phoneme string list by subtracting 1 from the phoneme number used in the fixed-length phoneme string extraction unit and the appearance from the storage unit after outputting the phoneme string list by the data deletion unit. Delete the frequency and the phoneme string and the appearance frequency for which the appearance frequency did not reach the threshold for the number of phonemes . As a result, phoneme strings other than the phoneme string necessary for the cumulative determination means to perform the determination are deleted, and the capacity of the storage means can be kept low.

Also, list generation apparatus for speech synthesis according to claim 2, a plurality of word text writing down the speech into a word of the text data, speech synthesis list generating apparatus for generating a word string list of the search unit of the speech synthesis a is provided with a fixed-length word string extracting unit, a frequency cumulative means, and storage means, and a list output unit, the occurrence frequency cumulative means cumulative meter determining means, said list output unit data deleting means It is characterized by.

  According to such a configuration, the speech synthesis list generating device uses the fixed-length word string extraction unit to extract words corresponding to the number of consecutive words in each word text from the word text so that the number of words is a predetermined number. Extract as a column. Here, the word text is a voice written in the text data of the word. For example, when the voice “Nationwide weather” is written in the word, the word text is “zenkoku / no / tenki”. / Desu "(" / "is a word break).

In addition, the speech synthesis list generation device uses the total determination unit to generate word string appearance frequencies corresponding to the number of words one shorter than the number of words in the word string from the beginning of the word string extracted by the fixed-length word string extraction unit. subject but if more than a predetermined threshold value, the word string extracted with fixed-length word string extraction unit, determines that the object to be cumulative occurrence frequency, the frequency accumulation means, for accumulating the frequency of occurrence The frequency of appearance of word strings determined to be is accumulated.
Then, the speech synthesis list generating device refers to the word string list in which the appearance frequency is higher than a predetermined threshold by referring to the storage means for storing the word string and the appearance frequency for each word count by the list output means. Output as.

  Note that the fixed-length word string extraction unit sequentially increases the number of words and extracts a word string from the word text, and the list output unit outputs word string lists having different numbers of words sequentially. As a result, the speech synthesis list generating device accumulates the appearance frequencies in order from the word string having the smallest number of words, and sequentially selects only word strings including word strings having a high appearance frequency (more than the threshold) as the accumulation target. Since word strings are accumulated, a word string list can be generated at high speed.

In addition, the speech synthesis list generation device outputs the word string list by the data deletion unit, and then stores the word sequence of the number of words by subtracting 1 from the number of words used in the fixed-length word sequence extraction unit and the appearance from the storage unit The frequency, the word string whose appearance frequency did not reach the threshold for the number of words, and the appearance frequency are deleted. As a result, word strings other than the word string necessary for the cumulative determination means to make a determination are deleted, and the capacity of the storage means can be kept low.

Furthermore, the speech synthesis list generation program according to claim 3 , the computer generates a phoneme string list as a speech synthesis search unit from a plurality of phoneme texts in which speech is written in phoneme text data. The fixed-length phoneme string extraction unit, the total determination unit, the appearance frequency totaling unit, and the list output unit function.

According to this configuration, the speech synthesis list generation program extracts a phoneme string composed of continuous phonemes having a predetermined number of phonemes from each phoneme text by the fixed-length phoneme string extraction means.
Then, the speech synthesis list generation program performs a phoneme equivalent to the number of phonemes one shorter than the number of phonemes of the phoneme sequence from the beginning of the phoneme sequence shorter than the phoneme sequence extracted by the fixed-length phoneme sequence extraction unit by the cumulative determination unit. When the appearance frequency of the sequence is higher than a predetermined threshold value, it is determined that the phoneme sequence extracted by the fixed-length phoneme sequence extraction unit is a target for accumulating the appearance frequency. Note that the phoneme string shorter than the phoneme string extracted by the fixed-length phoneme string extracting means and the appearance frequency thereof are those already stored in the storage device.

Then, the speech synthesis list generation program accumulates the appearance frequencies of the phoneme strings determined by the appearance frequency accumulating means by the accumulating determination means to be the object of accumulating the appearance frequencies, and adds the phoneme strings and the phoneme strings. The corresponding appearance frequency is stored in the storage device. As a result, only phoneme strings having a high appearance frequency are stored in the storage device.
As described above, at the stage where the phoneme string and its appearance frequency are stored in the storage device from the phoneme text, the speech synthesis list generation program outputs a phoneme string whose appearance frequency is higher than a predetermined threshold as a phoneme string list. .

The fixed-length phoneme string extraction means sequentially increases the number of phonemes to extract a phoneme string from the phoneme text, and the list output means outputs a phoneme string list having different phoneme numbers. As a result, the speech synthesis list generation program accumulates the appearance frequencies in order from the phoneme sequence having the smallest number of phonemes, and sequentially selects only the phoneme sequences including the phoneme sequences having the high appearance frequency (greater than the threshold value). Since the phoneme strings are accumulated, a phoneme string list can be generated at high speed.
In addition, the list generation program generates the phoneme sequence and the appearance frequency of the phoneme number obtained by subtracting 1 from the phoneme number used in the fixed-length phoneme sequence extraction unit from the storage unit, and the appearance number and the phoneme number appear by the list output unit Delete the phoneme string and the appearance frequency whose frequency did not reach the threshold. As a result, phoneme strings other than the phoneme string necessary for the cumulative determination means to perform the determination are deleted, and the capacity of the storage means can be kept low.

According to a fourth aspect of the present invention, there is provided a speech synthesis list generation program for generating a word string list as a speech synthesis search unit from a plurality of word texts in which speech is written in word text data. The fixed-length word string extraction unit, the total determination unit, the appearance frequency total unit, and the list output unit function.

According to this configuration, the speech synthesis list generation program extracts a word string composed of consecutive words of a predetermined number of words from each word text by the fixed-length word string extraction unit.
Then, the speech synthesis list generation program uses the cumulative determination means to start from the beginning of the word string that is shorter than the word string extracted by the fixed-length word string extraction means, and the number of words is one word shorter than the number of words in the word string. When the appearance frequency of the sequence is higher than a predetermined threshold, it is determined that the word sequence extracted by the fixed-length word sequence extraction unit is a target for accumulating the appearance frequency. Note that the word string shorter than the word string extracted by the fixed-length word string extracting unit and the appearance frequency thereof are those already stored in the storage device.

Then, the speech synthesis list generation program accumulates the appearance frequencies of the word strings determined by the appearance frequency accumulating means by the accumulating determination means to be the object of accumulating the appearance frequencies, and adds the word strings and the word strings. The corresponding appearance frequency is stored in the storage device. As a result, only word strings having a high appearance frequency are stored in the storage device.
In this way, at the stage where the word string and its appearance frequency are stored in the storage device from the word text, the speech synthesis list generation program outputs a word string whose appearance frequency is higher than a predetermined threshold as a word string list. .

Note that the fixed-length word string extraction unit sequentially increases the number of words and extracts a word string from the word text, and the list output unit outputs word string lists having different numbers of words sequentially. As a result, the speech synthesis list generation program accumulates the appearance frequencies in order from the word string having the smallest number of words, and sequentially selects only word strings including word strings having a high appearance frequency (greater than the threshold) as the accumulation target. Since word strings are accumulated, a word string list can be generated at high speed.
In addition, the speech synthesis list generation program causes the list output unit to generate the word sequence of the number of words obtained by subtracting 1 from the number of words used in the fixed-length word sequence extraction unit, the appearance frequency, and the number of words from the storage unit. Delete the word string and the appearance frequency whose frequency did not reach the threshold. As a result, word strings other than the word string necessary for the cumulative determination means to make a determination are deleted, and the capacity of the storage means can be kept low.

According to the invention described in claim 1 or claim 3 , the appearance frequencies of the phoneme strings are sequentially accumulated from the phoneme strings having a small number of phonemes, and the phoneme strings including the phoneme strings whose appearance frequencies are higher than a predetermined threshold are included. Since only the phoneme strings are subject to appearance frequency accumulation, waste of accumulating the appearance frequencies of phoneme strings that have been found to be low in appearance frequency can be eliminated, and a phoneme string list can be generated at high speed. At this time, since all phoneme strings higher than the threshold value can be extracted, the sound quality degradation at the connection point of the phoneme strings is reduced when the speech synthesis is performed using the phoneme string list generated by the present invention. It becomes possible.

Further, according to the invention described in claim 1 or claim 3, the phoneme sequence number of phonemes that accumulates the current frequency in the storage means, a higher number of phonemes than frequency reaches a predetermined threshold - Since only one phoneme string is stored, the capacity of the storage means can be kept low.

According to the invention described in claim 2 or claim 4 , the appearance frequency of the word string is sequentially accumulated from the word string having a small number of words, and the word string including the word string whose appearance frequency is higher than a predetermined threshold is included. Since only word strings are targeted for the appearance frequency accumulation, waste of accumulating the appearance frequencies of word strings that are known to have a low appearance frequency can be eliminated, and a word string list can be generated at high speed. At this time, since all word strings higher than the threshold can be extracted, deterioration of sound quality at the connection points of the word strings is reduced when performing speech synthesis using the word string list generated by the present invention. It becomes possible.

Further , according to the invention of claim 2 or claim 4 , the storage means includes a word string of the number of words currently accumulated in appearance frequency, and the number of words whose appearance frequency is higher than a predetermined threshold value − Since only one word string is stored, the capacity of the storage means can be kept low.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
[Outline of list generator for speech synthesis]
First, an outline of the speech synthesis list generation apparatus will be described with reference to FIGS. 3 and 4. FIG. 3 is an explanatory diagram for explaining a phoneme string list generated by the speech synthesis list generation apparatus. FIG. 4 is an explanatory diagram for explaining an outline of the operation of generating a phoneme string list in the speech synthesis list generating apparatus.

  As shown in FIG. 3, the speech synthesis list generation device 10 generates a phoneme sequence list (phoneme sequence, appearance frequency) that is a search unit for speech synthesis from a plurality of phoneme texts in which speech is written in phoneme text data. To do. Here, it is assumed that the speech synthesis list generation apparatus 10 inputs phoneme text from the speech database 20 in which phoneme text is stored, and outputs (writes) the generated phoneme string list to the speech synthesis database 30.

  For example, “ny / u: / s / u / d / e / s / u” (“/” is a phoneme break; the same applies hereinafter) is voiced as a phoneme text in which “news” is written in phoneme text data. When stored in the database 20, the speech synthesis list generation apparatus 10 accumulates the appearance frequencies of phoneme strings for each phoneme number in the speech database 20. Then, the speech synthesis list generation device 10, for example, synthesizes a value obtained by accumulating the appearance frequency of the phoneme string “u: / s / u” with the number of phonemes “3” (here “156”) together with the phoneme string. Output to the database 30 for use.

  Next, the outline of the operation of the speech synthesis list generation apparatus 10 will be described with reference to FIG. 4 (refer to FIG. 3 as appropriate). First, the speech synthesis list generator 10 generates a phoneme having a phoneme number of “3” as an initial phoneme number from a phoneme text (here “ny / u: / s / u / d / e / s / u”). Accumulate the appearance frequency of the column. Then, only the phoneme string in which the appearance frequency of the phoneme string exceeds a predetermined threshold (here, “100”) is output to the speech synthesis database 30. For example, “u: / s / u” which is the appearance frequency exceeding the threshold is output.

Then, the speech synthesis list generating apparatus 10 increases the number of phonemes by 1 (only the number of phonemes “4” in this case) for only the phoneme strings whose appearance frequency exceeds the threshold, and accumulates the appearance frequencies of the phoneme strings. . Here, as in the case of the number of phonemes “3”, only the phoneme string whose appearance frequency exceeds the threshold is output to the speech synthesis database 30. For example, (s / u / d / e) that is the appearance frequency exceeding the threshold is output.
In this way, the phoneme string that sequentially exceeds the appearance frequency while increasing the number of phonemes by 1 and the appearance frequency are output to the speech synthesis database 30.

Accordingly, a phoneme string longer than that of a phoneme string having a low appearance frequency can be omitted from the accumulation target of the appearance frequency, a memory capacity can be reduced, and a phoneme string list can be generated at high speed. Here, although the initial phoneme number is “3”, it may be started from “1”. Further, the list (phoneme string list) is generated in units of phonemes, but the list (word string list) may be generated in units of words.
When the present invention was applied to the speech database 20 in which about 30000 sentence phoneme texts were accumulated, the phoneme string appearance frequency threshold was set to “100”, about 40000 phoneme string lists were generated. As a result, it is possible to collectively search from the phoneme string list for a fixed phrase that depends on the speech database 20.
Hereinafter, the configuration and operation of the speech synthesis list generation device will be described in detail.

[Configuration of List Generation Device for Speech Synthesis]
First, the configuration of the speech synthesis list generation apparatus according to the present invention will be described with reference to FIG. FIG. 1 is a block diagram illustrating a configuration of a speech synthesis list generation apparatus. As shown in FIG. 1, the speech synthesis list generation apparatus 10 inputs a plurality of phoneme texts, generates a list of phoneme strings (phoneme string list) having an appearance frequency equal to or higher than a certain threshold, and outputs the list. . Here, the phoneme text is a phonetic text written into phoneme text data which is the minimum unit of speech.

  Here, the speech synthesis list generation apparatus 10 includes a fixed-length phoneme string extraction unit 11, an appearance frequency accumulation unit 12, a storage unit 13, a list output unit 14, and an initial data setting unit 15. . Also, here, the speech synthesis list generation device 10 includes a speech database 20 that stores phoneme text, a speech synthesis database 30 that stores the generated phoneme string list, and an input means 40 such as a keyboard. Are connected to the outside.

  The fixed-length phoneme string extraction means 11 extracts a phoneme string composed of continuous phonemes having a fixed-length phoneme number from the phoneme text. Here, the fixed-length phoneme sequence extraction unit 11 sequentially reads phoneme texts from the speech database 20, extracts a fixed-length phoneme sequence, and outputs it to the appearance frequency accumulation unit 12. The fixed-length phoneme string extraction unit 11 uses the initial phoneme number set by the initial data setting unit 15 as the initial value of the phoneme number, and extracts the phoneme string from all phoneme texts in the speech database 20 at the stage. The output unit 14 is notified of an output instruction including the number of phonemes. Then, the fixed-length phoneme string extraction unit 11 is notified of the completion of output from the list output unit 14, and continuously increases the number of phonemes by 1 and extracts a phoneme string.

  For example, “ny / u: / s / u / d / e / s / u” is input as the phoneme text in which “news” is written in the phoneme text data, and “3” is set as the initial phoneme number. In this case, the fixed-length phoneme string extraction means 11 has a plurality (six here) of phoneme strings “ny / u: / s”, “u: / s / u”, “s” composed of three phonemes. / U / d "," u / d / e "," d / e / s "and" e / s / u "are extracted. In addition, the fixed-length phoneme string extraction unit 11 extracts a phoneme string “ny / u: / s / u” or the like obtained by increasing the number of phonemes by being notified of the output completion from the list output unit 14.

The appearance frequency accumulating unit 12 inputs the phoneme strings extracted by the fixed-length phoneme string extracting unit 11 and accumulates the appearance frequencies of the phoneme strings. The appearance frequency accumulating unit 12 searches the storage unit 13, and when the input phoneme sequence is a new phoneme sequence (not stored in the storage unit 13), the phoneme sequence and its appearance are stored in the storage unit 13. A region for storing the frequency is secured, the input phoneme string is stored in the region, and the appearance frequency is initialized (for example, “1” is set). If the input phoneme string is already stored in the storage means 13, 1 is added to the appearance frequency corresponding to the phoneme string. As a result, the paired data of the phoneme string and the appearance frequency is stored in the storage unit 13.
Here, the appearance frequency accumulating means 12 is provided with a cumulative determination unit 12a, and it is determined whether or not the appearance frequencies of the input phoneme strings are to be accumulated.

  The total determination unit 12a (total determination means) determines whether the input phoneme string is a target phoneme string for which the appearance frequency is to be accumulated. The total determination unit 12a searches the storage unit 13 for the appearance frequency of a phoneme string shorter than the input phoneme string, and inputs the input frequency when the appearance frequency is equal to or greater than the threshold set by the initial data setting unit 15. The determined phoneme string is determined to be a target for accumulating the appearance frequency.

For example, in the phonetic text “ny / u: / s / u / d / e / s / u” of “It is news”, the phoneme string “u: / s / u”, “ When the appearance frequency of “s / u / d” and “d / e / s” is equal to or greater than a threshold value (for example, “100”), the cumulative determination unit 12a includes “u: / s / It is determined that “u / d”, “s / u / d / e”, and “d / e / s / u” are targets for accumulating appearance frequencies. That is, the cumulative determination unit 12a has a low appearance frequency of the phoneme strings “ny / u: / s”, “u / d / e”, and “e / s / u” composed of the other three phonemes. Since it is known that it is not meaningful to accumulate the appearance frequencies of phoneme strings having the above length, these are excluded from the appearance frequency accumulation targets.
Based on the determination result of the total determination unit 12a, the appearance frequency totaling unit 12 can limit phoneme strings for measuring the appearance frequency.

  The storage means 13 (storage device) temporarily stores a phoneme string and its appearance frequency, and is a general recording medium such as a memory. In the storage means 13, the phoneme string is written by the appearance frequency accumulating means 12, and the appearance frequency corresponding to the phoneme string is sequentially updated.

  The list output means 14 outputs a phoneme string having a high appearance frequency among the phoneme strings stored in the storage means 13 as a phoneme string list. The list output unit 14 is notified of an output instruction including the number of phonemes from the fixed-length phoneme sequence extraction unit 11, and selects a phoneme sequence having a high appearance frequency from among the phoneme sequences composed of the phoneme numbers. Output. Here, the list output unit 14 includes an output determination unit 14a and a data deletion unit 14b.

  The output determination unit 14 a determines whether to output the phoneme string stored in the storage unit 13. The output determination unit 14 a determines that a phoneme string having an appearance frequency equal to or higher than the threshold set by the initial data setting unit 15 among the phoneme strings stored in the storage unit 13 can be output. As a result, it is possible to output only phoneme strings that have an appearance frequency equal to or higher than a certain threshold as a phoneme string list.

  The data deletion unit 14b (data deletion unit) outputs a phoneme string having a high appearance frequency as a phoneme string list, and then deletes a phoneme string stored in the storage unit 13 and unnecessary data in the appearance frequency. Then, the storage (memory) area is released. The data deletion unit 14b is a phoneme sequence of phonemes obtained by subtracting 1 from the number of phonemes included in the output instruction notified from the fixed-length phoneme sequence extraction unit 11, and the number of phonemes included in the output instruction. The phoneme string whose appearance frequency has not reached the minimum number of data (threshold) and the data of the appearance frequency are deleted. Here, in the phoneme string of the number of phonemes included in the output instruction, the phoneme string that has reached the threshold value is left in the cumulative determination unit 12a, in which the phoneme string that is one shorter than the input phoneme string is accumulated in the appearance frequency. This is because it is used to determine the target. Thereby, the capacity of the storage means 13 can be suppressed.

The initial data setting means 15 sets the initial phoneme number given to the fixed-length phoneme string extraction means 11 and the threshold values given to the appearance frequency accumulating means 12 and the list output means 14 as initial data. Here, the initial data setting means 15 inputs initial data from the input means 40 such as a keyboard. For example, by setting the initial phoneme number as “3” as a predetermined phoneme number, the fixed-length phoneme string extraction unit 11 sequentially sets “4”, “5”... With the phoneme number “3” as an initial value. A phoneme string is extracted by increasing the number of phonemes. Further, for example, by setting the threshold value to “100”, the cumulative determination unit 12a of the appearance frequency accumulating unit 12 sets the appearance frequency of the phoneme string one shorter than the input phoneme string and the threshold value (“100”). And the phoneme string input when the appearance frequency is equal to or higher than the threshold value is determined to be a target for accumulating the appearance frequency. The list output unit 14 outputs a phoneme string having an appearance frequency equal to or higher than a threshold value to the outside.
When the initial phoneme number and the threshold value are set to specific values set in advance, the initial data setting unit 15 may be removed from the speech synthesis list generation device 10.

  As described above, the configuration of the speech synthesis list generation device 10 has been described based on one embodiment, but the speech synthesis list generation device 10 uses a fixed-length phoneme string extraction unit 11, an appearance frequency accumulation unit 12, and a list output in a computer. The means 14 can be realized as a function program, and the function programs can be combined to operate as a speech synthesis list generation program.

  In addition, here, the speech synthesis list generation apparatus 10 has been described as generating a phoneme string list, but may function as a word string list. For example, “zenkoku / no / tenki / desu” (“/” is a word delimiter), which is a word text in which “National weather” is written in the text data of the word, is input based on the appearance frequency for each word string It is also possible to generate a word string list.

  In this case, as shown in FIG. 2, the fixed-length phoneme string extraction means 11 of FIG. 1 extracts a word string composed of continuous words having a fixed number of words from the word text. Function as 11B. Further, the appearance frequency accumulating means 12, the accumulating means 13 and the list output means 14 shown in FIG. 1 are accumulated, accumulated and outputted in units of word strings instead of phoneme strings, the appearance frequency accumulating means 12B, the accumulating means 13B and the list output. It functions as the means 14B. Further, the initial data setting means 15 of FIG. 1 is caused to function as initial data setting means 15B for setting the initial word number instead of the initial phoneme number.

  By configuring the speech synthesis list generation device 10B as shown in FIG. 2, the speech synthesis list generation device 10B inputs a plurality of word texts, and a list of word strings having an appearance frequency equal to or higher than a certain threshold ( A word string list) can be generated and output. 2 has the same function as the speech synthesis list generation device 10 except for the use of a word string instead of a phoneme string, and a detailed description thereof will be omitted. .

[Operation of List Generation Device for Speech Synthesis]
Next, the operation of the speech synthesis list generation apparatus according to the present invention will be described with reference to FIG. FIG. 5 is a flowchart showing the operation of the speech synthesis list generation apparatus. Here, it is assumed that the initial data setting means 15 sets the initial phoneme number, which is the initial value of the phoneme number sequence, and the appearance frequency threshold value in advance.

(Fixed long phoneme string extraction step)
First, the speech synthesis list generation apparatus 10 inputs one phoneme text from the speech database 20 by the fixed-length phoneme string extraction means 11 (step S1), and then consists of phonemes composed of continuous phonemes of the set number of phonemes. A column is extracted (step S2). For example, from the phoneme text “ny / u: / s / u / d / e / s / u”, a phoneme string with a set phoneme number (in the case of the phoneme number “3”, “ny / u: / s”, etc. ).

(Cumulative judgment step)
Then, the cumulative determination unit 12a of the appearance frequency accumulating unit 12 determines whether the phoneme string is a target phoneme string for which the appearance frequency is accumulated (step S3). Here, the appearance frequency of the phoneme string one shorter than the input phoneme string is searched from the storage means 13, and when the appearance frequency is equal to or higher than the threshold set by the initial data setting means 15, the input phoneme string is Is determined to be a target for accumulating the appearance frequency.

(Appearance frequency accumulation step)
Here, if it is determined that the phoneme string is an object of the appearance frequency (Yes in step S3), the appearance frequency accumulating unit 12 increases the appearance frequency of the phoneme string by 1 (step S4), and the process proceeds to step S5. If the phoneme string is new, an area for storing a new phoneme string and the appearance frequency of the phoneme string is secured in the storage unit 13. On the other hand, when it is determined that the phoneme string is not the target of the appearance frequency (No in step S3), the process proceeds to step S5 as it is.

  Then, the speech synthesis list generation device 10 determines whether or not the phoneme string has been extracted from the phoneme text input in step S1 (step S5). That is, it is determined whether or not a phoneme string having the number of phonemes set from the phoneme text can be newly extracted. If not all phoneme strings have been extracted from the phoneme text (No in step S5), the process returns to step S2 to cut out the phoneme string next to the set phoneme number from the phoneme text and continue the operation. .

  Further, when all the phoneme strings have been extracted (Yes in step S5), the speech synthesis list generation device 10 further determines whether or not the input from the speech database 20 has been completed (step S6). Here, when the input from the voice database 20 is completed (Yes in step S6), the process proceeds to step S7. On the other hand, if the input from the speech database has not been completed (No in step S6), that is, if there is a phoneme text that has not yet been extracted from the phoneme string, the process returns to step S1 and the speech is returned. Enter phoneme text from database and continue operation.

(List output step)
Then, the speech synthesis list generation device 10 notifies the list output unit 14 of an output instruction from the fixed-length phoneme sequence extraction unit 11, and the list output unit 14 includes the phoneme sequence stored in the storage unit 13. The phoneme strings having an appearance frequency equal to or higher than the threshold set by the initial data setting unit 15 are output as a phoneme string list (step S7). Here, the data deletion unit 14b deletes the phoneme sequence of the number of phonemes obtained by subtracting 1 from the number of phonemes included in the output instruction and the data of the appearance frequency from the storage unit 13 (not shown as steps). ).

Thereafter, when the output completion is notified from the list output means 14, the fixed-length phoneme string extraction means 11 increases the number of phonemes by 1 (step S8), and a phoneme string equal to or greater than the number of phonemes is present in the speech database 20. If it exists (Yes in step S9), the process returns to step S1, inputs the phoneme text from the speech database, and continues the operation. If no phoneme string equal to or greater than the number of phonemes is present in the speech database 20 (No in step S9), the operation is terminated.
Through the above operation, the speech synthesis list generation apparatus 10 can input a plurality of phoneme texts and output a list of phoneme strings (phoneme string list) having an appearance frequency equal to or higher than a certain threshold.

It is a block diagram which shows the structure of the list production | generation apparatus for speech synthesis which produces | generates the phoneme string list which concerns on this invention. It is a block diagram which shows the structure of the list production | generation apparatus for speech synthesis which produces | generates the word string list which concerns on this invention. It is explanatory drawing for demonstrating the outline | summary of the phoneme string list | wrist produced | generated with the list | wrist production | generation apparatus for speech synthesis which concerns on this invention. It is explanatory drawing for demonstrating the outline | summary of operation | movement of the list production | generation apparatus for speech synthesis which concerns on this invention. It is a flowchart which shows operation | movement of the list production | generation apparatus for speech synthesis which concerns on this invention.

Explanation of symbols

10, 10B List generator for speech synthesis 11 Fixed-length phoneme string extraction means 11B Fixed-length word string extraction means 12, 12B Appearance frequency accumulation means 12a Accumulation determination unit (accumulation determination means)
13, 13B Storage means (storage device)
14, 14B List output means 14a Output determination part 14b Data deletion part (data deletion means)
15, 15B Initial data setting means

Claims (4)

A speech synthesis list generation device that generates a phoneme string list that is a search unit for speech synthesis from a plurality of phoneme texts in which speech is written into phoneme text data,
A fixed-length phoneme string extraction means for sequentially increasing a predetermined number of phonemes and sequentially extracting a phoneme string composed of continuous phonemes of the number of phonemes from each of the phoneme texts;
Appearance frequency accumulating means for accumulating the appearance frequencies at which the phoneme strings extracted by the fixed-length phoneme string extraction means appear in the plurality of phoneme texts;
Accumulating means for accumulating the phoneme string and the appearance frequency of the phoneme string accumulated by the appearance frequency accumulating means;
A list output that outputs, as the phoneme sequence list, a phoneme sequence in which the frequency of appearance of phoneme sequences stored in the storage unit is greater than a predetermined threshold each time the number of phonemes is increased by the fixed-length phoneme sequence extraction unit Means and
The appearance frequency accumulating means extracts the extracted phoneme string when the appearance frequency of the phoneme string one number shorter than the phoneme number of the phoneme string from the head of the extracted phoneme string is higher than the threshold . A cumulative determination unit that determines that a phoneme string is a target of cumulative appearance frequency ;
After outputting the phoneme string list, the list output means outputs a phoneme string and an appearance frequency of a phoneme number obtained by subtracting 1 from the phoneme number used in the fixed-length phoneme string extraction means from the storage means, A speech synthesizing list generation apparatus comprising: a data deletion unit that deletes a phoneme string and an appearance frequency that are prime numbers whose appearance frequency has not reached the threshold value . A speech synthesis list generation device that generates a word string list, which is a search unit for speech synthesis, from a plurality of word texts in which speech is written into word text data,
A fixed-length word string extraction means for sequentially increasing a predetermined number of words and sequentially extracting a word string composed of words having a continuous number of words from each of the word texts;
Appearance frequency accumulating means for accumulating the appearance frequencies of the word strings extracted by the fixed-length word string extracting means,
Accumulating means for accumulating the word string and the appearance frequency of the word string accumulated by the appearance frequency accumulating means;
Each time the number of words is increased by the fixed-length word string extraction means, a list output that outputs as a word string list a word string in which the appearance frequency of the word strings stored in the storage means is higher than a predetermined threshold Means and
The appearance frequency accumulating means extracts the extracted word string when the appearance frequency of the word string corresponding to the number of words shorter by one than the number of words in the word string is higher than the threshold from the beginning of the extracted word string . A cumulative determination unit that determines that a word string is a target of cumulative appearance frequency ;
In the list output means, after outputting the word string list, a word string of the number of words obtained by subtracting 1 from the number of words used in the fixed-length word string extraction means and the appearance frequency from the storage means, and the word A speech synthesizing list generating apparatus comprising data deletion means for deleting a word string whose appearance frequency does not reach the threshold and the appearance frequency . In order to generate a phoneme string list as a search unit for speech synthesis from a plurality of phoneme texts in which speech is written into phoneme text data,
A fixed-length phoneme string extraction means for sequentially increasing a predetermined number of phonemes and sequentially extracting a phoneme string composed of continuous phonemes of the phoneme number from each of the phoneme texts;
If the frequency of appearance of phoneme strings corresponding to the number of phonemes one shorter than the number of phonemes in the phoneme string from the beginning of the phoneme string extracted by the fixed-length phoneme string extraction unit is greater than a predetermined threshold, the extracted phoneme string is extracted. A cumulative determination unit that determines that the phoneme string is a target of cumulative appearance frequency;
Appearance frequency accumulating means for accumulating the appearance frequencies of the phoneme strings determined to be the object of appearance frequency accumulation by the accumulative determination means, and accumulating the phoneme strings and the appearance frequencies corresponding to the phoneme strings in the storage device;
Each time the number of phonemes is increased by the fixed-length phoneme string extraction means, a phoneme string in which the appearance frequency of the phoneme strings stored in the storage device is greater than the threshold is output as the phoneme string list , and After outputting the phoneme sequence list, the phoneme sequence and the appearance frequency of the phoneme number obtained by subtracting 1 from the phoneme number used by the fixed-length phoneme sequence extraction unit from the storage unit, and the appearance frequency at the phoneme number are set as the threshold value. List output means for deleting phoneme strings and appearance frequencies that have not been reached ,
A program for generating a list for speech synthesis, characterized in that In order to generate a word string list, which is a search unit for speech synthesis, from a plurality of word texts in which speech is written into word text data,
A fixed-length word string extraction means for sequentially increasing a predetermined number of words and sequentially extracting a word string composed of words having a continuous number of words from each of the word texts;
If the appearance frequency of word strings corresponding to the number of words one shorter than the number of words in the word string from the beginning of the word string extracted by the fixed-length word string extraction unit is greater than a predetermined threshold, the word string is extracted. A cumulative determination means for determining that the word string is a target of cumulative appearance frequency,
Appearance frequency accumulating means for accumulating the appearance frequency of the word string determined to be the object of appearance frequency accumulation by the accumulative determination means, and accumulating the word string and the appearance frequency corresponding to the word string in the storage device,
Each time the number of words is increased by the fixed-length word string extraction unit, a word string in which the frequency of occurrence of the word string stored in the storage device is greater than the threshold is output as the word string list , and After outputting the word string list, from the storage means, the word string of the number of words obtained by subtracting 1 from the number of words used in the fixed-length word string extraction means and the appearance frequency, and the appearance frequency with the number of words is set as the threshold value. A list output means for deleting the word string and the appearance frequency that have not been reached ;
A program for generating a list for speech synthesis, characterized in that

Images