JPS6349243B2 - - Google Patents

Description

[Detailed Description of the Invention] "Industrial Application Field" This invention relates to an audio output device that converts Japanese sentences written in character codes into audio, and particularly relates to a unit that sets a pause section in the output audio. It is.

"Prior Art" The configuration of the conventional audio output device shown in FIG. 1 is shown. The Japanese text written in the character code from the input terminal 11 is passed to the language processing section 12. Language processing unit 1
In step 2, the kanji, alphanumeric characters, and symbols in the input Japanese text are read out from memory according to their character codes and converted into pronunciations, and at the same time, the accent position and voice pause interval are specified. Speech synthesis section 13
Then, audio parameters are combined based on information such as pronunciations, accents, audio pause sections, etc. from the language processing unit 12, and audio is output from the audio output unit 14.

In order to improve the naturalness and intelligibility of the voice output from this voice output device, it is necessary to set appropriate voice pauses in the output voice. However, with conventional devices of this type, speech pauses are only set at very limited positions such as punctuation marks, and it is difficult to automatically define natural speech pauses in input Japanese sentences that are not separated. . As a result, the output audio has fewer audio pauses, which is one of the major causes of deterioration in naturalness and intelligibility.

"Purpose of the Invention" The purpose of this invention is to convert an input Japanese sentence that is not separated into speech into speech, and to automatically insert natural pauses into the speech. An object of the present invention is to provide an audio output device capable of outputting audio.

"Structure of the Invention" According to the present invention, there is a punctuation mark detection unit that detects punctuation marks from input Japanese text, a symbol detection unit, a particle “wa” detection unit, a number detection unit, a compound word boundary detection unit, and a particle detection unit. and/or a character type transition point detection unit that detects a transition point from hiragana to a character type other than hiragana, and a pause interval determination unit. An audio pause section (pause) is inserted into the output audio.

Embodiment FIG. 2 shows an example of a language processing unit which is a main part of the audio output device according to the present invention. In FIG. 2, a Japanese sentence inputted from the input terminal 11 is converted into a reading and an accent in the accent adding unit 15, for example, a reading and writing accent table is read out based on each input character code. The signal is then accented and supplied to the terminal 16 for input to the speech synthesis section 13.

In addition, a symbol detection unit 17,
A symbol, a particle "ha", and a punctuation mark are detected by the particle "wa" detection unit 18 and the punctuation mark detection unit 19, respectively. When these detecting units 17, 18, and 19 detect the corresponding one, the pause interval setting unit 21 creates a pause (pause) of an appropriate length immediately after the symbol, the particle "ha", or the punctuation mark. The reading accent information obtained by the reading accent adding section 15 is sent to the speech synthesis section 13 through the terminal 16.

When the number detection unit 22 detects a number string from the input Japanese text, a pause section is set immediately before the number string and between every four digits of 10,000, 100,000,000, and trillion.

For example, if "1234567890" is entered, △12△
3456△7890 (△Jiyounioku△Sanzenyonhyakugojieurokuman△Nanasenhatsupiyakukiyuu). Here, △ indicates a pause section.

When reading a string of numbers digit by digit, such as an account number, read 4 from the beginning (immediately) and the beginning of the string of numbers.
Set a pause interval for each character.

Example "1234567890" → △1234△5678△90 (△Ichini Sanyeon△Goroku Nanahachi△Kyuu Zero) When the particle detection unit 23 detects a particle from the input Japanese sentence, the cumulative number of moras m from the previous pause section is If M ₀ ≦m≦M ₁ , a pause section is set immediately after the particle. At this time, the cumulative number m of moras from the immediately preceding rest period is counted by a mora number counter 24, and each time a rest period is set by the rest period setting section 21, this mora number counter 24 is reset.

Also, the cumulative mora number m of the counter 24 is m>M ₁
If so, the particle detected by the particle detection unit 23, the point at which hiragana changes to a character type other than hiragana detected by the character type transition point detection unit 25, or the point detected by the compound word detection unit 26. A speech pause section is set immediately after the position where the word boundary in the compound word first appears. These determinations are made by the rest section determination unit 2.
Do it in 7.

Note that all word boundaries within compound words are not considered as candidates.
By selecting only accent phrase boundaries (word boundaries with accent breaks) within compound words as candidates, the effect becomes greater.

Example: Japan/Telegraph/Telephone/Public Corporation/Telecommunications/Research Institute, Tazushi ``/'' is an accented phrase boundary, and ``・'' is a word boundary. It is desirable that the judgment thresholds M ₀ and M ₁ shown here be set to be the length of the physiological pause interval when a human vocalizes (around 20 moras).
M ₀ is the minimum number of moras that require a pause section,
M ₁ is the maximum number of moras that does not require a rest section,
For example, M ₀ =5 and M ₁ =25, which differ depending on the detection conditions and text. Note that various detection units that have been conventionally used can be used.

The above audio pause interval setting method and setting position are summarized below.

(1) Pause immediately after the symbol (2) Pause immediately after the particle ``wa'' or punctuation mark (3) Immediately before the number string and immediately after the units of ten thousand, billion, trillion (in case of place reading), or 4 from the beginning Pause for each character (for single digit reading) (4) The cumulative number of moras from the previous pause section is M ₀ ≦
When m≦M ₁ , pause immediately before the particle (5) When the cumulative number of moras is greater than M ₁ , with the particle,
A pause immediately after the first occurrence of the transition point from hiragana to a character type other than hiragana and the word boundary within a compound word.The pause interval is determined using all the information in the detection section in Figure 2. An example of the processing operation to be set will be explained with reference to the flowchart in FIG. When a Japanese sentence is input in step _S1 , the mora number counter 24 is cleared in step _S2 , and m=0. In step _S3 , one character is read from the input Japanese text, and in step _S4 , the punctuation mark detection section 19 determines whether it is a punctuation mark.If it is determined to be a punctuation mark, a pause section is placed in the punctuation mark part in step _S5 . Set. Step S ₄
If it is determined that it is not a punctuation mark, it is determined in step S ₆ whether or not it is a symbol. If it is determined that it is a symbol, the reading and accent are specified in step S ₇ by the reading accent marking section 15, and further in step S ₈ . Set a pause section immediately after that symbol. If it is determined in step S ₆ that it is not a symbol, in step S ₉ the read character is a particle "wa", but it is not a symbol.
If the particle is determined to be "wa", the process moves to step _S8 ; if the particle is not "wa", the particle detection unit ₂₃ determines whether it is another particle in step S10. If it is determined to be a particle, in step _S9 , the count value m of the mora number counter 24 at that time becomes _M0 .
_The rest interval determination unit ₂₇ determines whether or _not the _interval is between M _{1 and} If not, the mora number counter 24 is incremented by one step in step _S12 .

If it is determined in step _S10 that it is not a particle, the number detection unit 22 determines whether the read character is a number in step _S13 , and if a number is detected, a pause section is inserted immediately before the number string in step _S14 . and set
Read all consecutive number strings in step _S15 ,
Check the digit reading in step _S16 , and if it is a digit reading, set the digit reading and accent in step _S17 ,
A pause section will be set for every 10,000,000,000,000,000,000 trillion. If the read number does not have a digit reading, the reading/accent is set for each digit in step _S18 , and a pause interval is set for every four characters from the beginning of the number string.

If it is determined in step S ₁₃ that it is not a number, it is determined in step S ₁₉ whether the read character is a kanji, and if it is a kanji, all consecutive kanji strings are read in step S ₂₀ , and then word by word is read in step S ₂₁ . Set the pronunciation and accent and move on to step S ₁₂ .
The mora number counter 24 is incremented by the number of moras. In step _S22 , it is determined whether it is a particle, and if it is not a particle, it is checked in step _S23 whether there is a word boundary within the compound word, and if not, in step _S24 , whether the character type has been changed from hiragana to a character type other than hiragana. Check. If it is determined in these steps S ₂₂ , S ₂₃ , and S ₂₄ that the character is a particle, a word boundary, and has a transition to a character type other than hiragana, it is determined in step S ₂₅ whether the number of moras m is greater than or equal to M ₁ . However, if M is greater than or equal to ₁ , a pause section is set in step _S26 .

After setting the pause sections in steps S ₅ , S ₈ , S ₁₇ , S ₁₈ and S ₂₆ , the process moves to step S ₂₇ .
The mora number counter 24 is cleared. step
In _S28 , it is determined whether the sentence is the end, and if it is not the end, the process returns to step _S3 , and if it is the end of the sentence, the process ends. If it is determined in step _S24 that the level has changed from flat to flat, the process moves to step _S23 , and if it is determined in step _S25 that the number of moras m is not greater than _M1 , the process moves to step _S28 .

When setting a pause section in the above, its length depends on the conditions for inserting the pause section, for example.
An appropriate value between 300mS and 1000mS is set.
Next, an example of setting the pause section will be shown. First, the input Japanese sentence is shown, and then the input information to the speech synthesis unit 13 is shown, P _n (m=1, 2,...) shows the pause position and the pause section length, and "," shows the accent position. show.

Example 1 (Punctuation mark, pause interval setting by particle "wa") This is the head office of the Telegraph and Telephone Public Corporation.

Kochiraha P ₁ Denden Koushiya P ₂ Honshades P ₃ Examples 2 (Setting the pause section using numbers and symbols) 961234+12456=... Ki'age Eurokuma'bP ₄ Sennihyaku'
SA'NJYUYO'MP'RAS P ₅ ICHIMAN P ₄ NISE'NYO'NHYAKGOJIYUROK'KO'IKO'R _P5 ...Example 3 (Setting a pause interval based on a word boundary within a compound word) Example 4 (Setting a pause interval based on particles and character type transition points) ``Effects of the Invention'' As explained above, according to the present invention, when converting Japanese sentences written in character codes into audio, natural audio pauses can be automatically set not only for punctuation marks but also for other parts. Since it is an output device, it has the following advantages.

(1) The device can be simplified because it does not require complicated semantic analysis of Japanese.

(2) The naturalness and intelligibility of speech are improved.

Note that only a part of each detection unit explained in FIG. 2 is used, or the threshold value M ₀ , M ₁
By setting the above, it is possible to create an audio output device that can flexibly set audio pause sections depending on the content of the text and the application form. Depending on the audio output device, in addition to setting the pause section based on punctuation mark detection, the pause section may also be set based on at least one of steps S ₆ , S ₈ , S ₁₁ , S ₁₇ , S ₁₈ , and S ₂₅ in FIG. It may also be possible to only set the break period.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an example of the configuration of a conventional audio output device, FIG. 2 is a block diagram showing a specific example of the main language processing section of the audio output device according to the present invention, and FIG. 3 is a pause section setting. 3 is a flowchart illustrating an example of processing operation.

Claims (1)

[Claims] 1. In an audio output device that converts Japanese sentences containing character codes such as kanji, hiragana, katakana, alphanumeric characters, symbols, etc. into speech, a symbol detection unit that detects symbols and the particle "wa" are detected. a particle detection unit that detects a particle “wa”, a number detection unit that detects a number, a punctuation detection unit that detects a punctuation mark, the symbol detection unit, the particle “ha” detection unit, the number detection unit, and the punctuation detection unit. A pause section setting section that inserts a speech pause section into the output audio based on each detection output, a particle detection section that detects a particle, and a character type transition point detection section that detects the point at which hiragana changes to a character type other than hiragana. , a word boundary detection unit within a compound word that detects a word boundary within a compound word, a mora number counter that calculates the number of moras of an input Japanese sentence, the particle detection unit, the character type transition point detection unit, and the compound word. The audio is characterized by comprising: a pause section determining section that determines the number of moras of the mora number counter based on the detection output of the boundary detecting section and determines whether or not to set a pause section in the pause section setting section. Output device.