TW451183B - Coarticulation processing apparatus for Chinese speech synthesis - Google Patents

Abstract

By analyzing the situation of coarticulation of the adjacent syllables in a word string, the present invention can obtain some variation rules for coarticulation in which the preceding syllable is different as the initial phoneme state and tone of the succeeding syllable is different. Using these variation rules, a CV-VC(VV) coarticulation processing apparatus for handling coarticulation is established. The coarticulation processing apparatus finds the VC(VV) coarticulation segment by the combining relationship between the consonants and vowels of the preceding and succeeding syllables, and implements the overlapping and adding operation of waveforms between the preceding and succeeding syllables, so as to achieve a natural and fluent effect on the synthesis of speech.

Description

Employees ’cooperation of the Intellectual Property Bureau of the Ministry of Economic Affairs, printed by Du 451 1 83 A7 ----------- B7___ V. Description of the invention (1) Background of the invention The invention of the jaw field The present invention is related to the continuous processing of Chinese speech synthesis Device, in particular a continuous tone processing device for smooth transition from one syllable to the next syllable in Chinese speech synthesis. Related Technology Overview In Chinese speech synthesis, in order to make a series of syllables sound more fluent and smooth, it is necessary to perform a smoothing process on the syllable connection. This is called liaison processing. During Chinese speech synthesis, 'in order to obtain a smooth transition from one syllable in a character string to the next syllable of the character string, a liaison process that overlaps some phonemes of the previous syllable and some phonemes of the next syllable is necessary. of. Figure 3 shows the wideband spectrum of "Chinese" (Chinese) sound. It can be clearly seen from Fig. 3 that there is a liaison phenomenon. However, most traditional Chinese speech synthesis systems simply connect adjacent phonemes of two syllables in the “Chinese” string without considering consonant processing, as shown in Figure 4. This results in unnatural synthesized speech. Furthermore, the liaison processing technology used in the learned Chinese speech synthesis system is to simulate the time domain of the liaison segments of the character string, that is, to find the most suitable liaison in the huge liaison recording data first. Sound segment. An internal difference is then made between the previous syllable and the next syllable of the most suitable legato. The key to the above procedure is the decision of the best legato segment and finding the best legato segment from the recorded speech segment audio data. A paper entitled "Speech Synthesis of Chinese Suyin Two Words" was presented at the Ninth Symposium on Computational Linguistics in the Republic of China (1996), which is hereby incorporated by reference. (Fill in this page again) Install -------- Order ------- 1 * line

ΑΓ B7 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs. 5. Description of the Invention (2), and the reference "... Figure 3" 丨 〇〇 is a turn-by-turn list of Pinyin sentences that allows users to input the Pinyin sentences to be synthesized. 110 is a character string storage unit for storing a large amount of recorded audio data. It is a monophonic unit that is used to store monophonic voice recordings. It is to search for and analyze the found characters from the character string storage unit 110 based on the character string inputted by the Pinyin sentence to do the liaison processing to determine the liaison segment branch character string search unit. ⑽ is used to find the midpoint detection unit at the midpoint of the consonant section in the 疋 string. It is just a calculation unit used to calculate the sound length of a legato. 15 {) is based on the input Pinyin sentence, and detects the monophonic storage unit 18G to find the first syllable of the monophonic recorded speech data, and synthesizes the single g of recorded speech data to the previous syllable synthesis unit. 160 is a continuous-segment synthesizing unit that synthesizes the synthesized speech data output by the single-segment monophonic synthesis unit 50 and the continuous-segment synthesis. 17〇 is based on the input pinyin sentence, detects the monophonic storage unit 180 to find the last syllable of the monophonic record ^ speech data, and synthesizes the synthesized speech data output by the monophonic segment synthesizing unit 16 and the monophonic recording found. Back-end syllable synthesis unit that synthesizes back-end syllables of speech data. 190 is a synthesized speech output unit that outputs the synthesized speech data output result in the form of speech. It can be clearly seen from FIG. 5 that in the conventional Chinese speech synthesis system, the best consonant segment is detected from the character string storage unit 110 and the best monophonic speech data is detected from the monophonic storage unit 180. The detected data is synthesized to improve the naturalness and understanding of the synthesized speech output. For example, if you want to use the system shown in Figure 5 to synthesize a string of characters that use liaison processing, · Chinese (Chinese)), the user first inputs the characters in a phase · phonic way via the input V unit 1 10 String-Chinese then 'from the character I ^ ------— ^ (Please read the notes on the back before filling this page) This paper size is applicable to the national standard (CXS) A. 丨 size X 297 (45) 183

Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs

The string storage unit 110 finds a character string voice data record that matches the character string "Chinese", and there is a character string in the substring storage unit 110. The Chinese character string 3 The meta string detection unit 120 will find the character string voice data records that match the year string "Chinese" from the character string storage list Fan 10. Then, the speech data of the found character string is analyzed to determine the super 9 segments of the character string "Chinese", and the midpoint of the hyphenated segment of the character string "Chinese," is calculated in the midpoint detection unit 130. The calculation unit 140 calculates the length of the syllables of the Chinese character string “.”. The syllable synthesizing unit 150 detects the monophonic storage unit 180 to find a single-syllable voice data record that matches the word. Synthetic syllable synthesis The unit 160 synthesizes the monophonic speech data records and liaison segments found in the symbol 0. Then, the & The voice data record is combined with the synthesized voice data output by the segment synthesis unit 160 and a single-tone voice data record that matches the word "text". Finally, the synthesized voice output unit 90 outputs the synthesized voice data in the form of speech. However, if the character string storage unit does not match the character string "Xinwen," the character string data records will be based on the character string "Chinese, the vowel (乂 厶) of the preceding syllable, and the character string" In Chinese, the first syllable of the next syllable (乂 卜) is used to find the most similar liaison, for example: "Common question, (Six 乂 厶 乂,) is synthesized by the above method. This can cause very unnatural synthetic speech. Furthermore, the above-mentioned system requires about 55 MB of memory space to store a large number of character string voice data records, thereby wasting useful memory space. In addition, because the voice data record is used as the basic unit of synthesis, the frequency and sound length cannot be changed, and the detection and synthesis of the voice data record is time-consuming. (Please read the note f on the back before filling this page.) -------- Order --------- line. This paper size is applicable to China National Standard (CNS) A4 specifications (2 Sichuan X297 public love) 6 Α7 Β7 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 5. Description of the invention (4 To sum up, the above-mentioned conventional technologies have the following disadvantages: 1. A large amount of single-tone voice data records and character string voices must be stored Data record 2. If the desired character string voice data record is not in the character string 单元 unit, it will cause very unnatural synthesized speech. 3. Because the voice data record is used, it is impossible to change the sound length and order 4. It is very time-consuming to check out the voice data record. SUMMARY OF THE INVENTION Therefore, the basic object of the present invention is to provide a liaison processing device for smooth transition from one syllable to the next syllable in Chinese speech synthesis. The liaison processing device The shortcomings of the above-mentioned conventional technology can be solved. According to the present invention, a liaison processing device for Chinese speech synthesis includes: a dictionary memory storing a dictionary including a large number of Chinese character strings and its corresponding pinyin marks; Storage units for the start and end positions of Chinese syllables and connected syllables, pinyin marks corresponding to various Chinese syllables and connected syllables, and the start and end positions of the consonants and vowels of various Chinese syllables and connected syllables; stored in dictionary memory according to The dictionary analysis input eight pinyin sentences, cut this sentence into multiple character word analysis units. The sub-strings detected by the k3 analysis unit, which is determined by the storage unit-the character string must go through a liaison program in order to detect the word Does the string need to go through a liaison processing program: and in the input Pinyin sentence, the sound of each syllable segment of the 'string detected' ^ (CNS) A4 '.'ίο x 297 ^ ----- ----

------------ * Earthquake ------- Order --------- (Please read the notes on the back before filling in this page J 451 1 83 A7 B7 Economy Printed by the Ministry of Intellectual Property Bureau's Consumer Cooperatives V. Invention Description (5)

Speech synthesizing unit D that interpolates between segments to generate synthesized speech In the above-mentioned liaison processing device, the liaison segment stored in the storage unit is the initial sound of the Chinese character string subsyllable, as defined in Figure 6 . The CV-VC (VV) continuous speech processing device for Chinese speech synthesis of the present invention has the above-mentioned structure. First, a phonetic sentence entered by a user in the dictionary memory stored in the dictionary memory is a plurality of character strings. The syllable analysis unit then determines which syllables must be processed by a liaison program. Next, the base period of each syllable and the start and end positions of the consonants are detected from the syllable data storage unit. Finally, the speech synthesis unit calculates the sound length and audio information, calculates the change in sound length and audio, and synthesizes and outputs speech. BRIEF DESCRIPTION OF THE DRAWINGS The above and other objects and features of the present invention are more easily understood when referring to the corresponding reference numerals in the accompanying drawings when describing the following preferred embodiments. These diagrams are as follows: Figure 1 is used in accordance with the present invention. Square diagram of the Chinese speech synthesis and continuous speech processing device of the preferred embodiment. Fig. 2 shows a graph of syllable contents stored in the register unit 13 shown in Fig. 1. Figure 3 shows the broadband spectrum of "Chinese" spoken by a person. Figure 4 shows the broadband spectrum of "Chinese," voiced by a traditional Chinese speech synthesis system. Figure 5 is a system block diagram of a traditional Chinese speech synthesis system. Figure 6 shows the chart used in the preferred embodiment to determine whether the character string needs to pass through the syllable start syllable type. This paper size is applicable to the Chinese National Standard (CNS) A4 specification (Ox 297).- ---------- ^ ---- I --- Order --------- line *. ≪ Please read the notes on the back of tl before filling in this page) Ministry of Economy Wisdom A7 printed by the Consumer Cooperative of the Property Bureau _____ Β7 ___ 5. Description of the Invention (6) Figure 7 explains the explanation of the "Taiwan" string string processing according to the preferred embodiment of the present invention. Detailed description of the preferred embodiment The preferred embodiment according to the present invention will be described below with reference to related diagrams. FIG. 1 is a system block diagram of a tone processing device for Chinese speech synthesis according to an embodiment of the present invention. Referring to FIG. 1, an input unit 10 For example, a keyboard is included, and the user can directly input pinyin for speech synthesis Sentence. Character analysis unit 丨 丨 Analyze the input sentence according to the dictionary stored in the dictionary memory, body (storage unit) 12 that stores a large number of character strings and their corresponding pinyin marks, and divide the sentence into multiple character strings. And the position of the character is marked between the next two character strings. The syllable division unit 14 is based on the base week data storage unit 15 of the CV syllable and the VC (VV) syllable, and the VC syllable and VC (VV) syllable The tag data storage unit 16 determines which character must be subjected to liaison processing, and finds out the base period data and mark data of CV syllables and VC. (VV) liaison segments that are determined to require liaison processing. Pitch length detection unit 1 7 and the audio detection unit 1 8 detect the corresponding sound length and audio according to the syllable prosody rules. Among them, C represents the child factor, and V represents the mother. Td The temporary storage area (storage) unit 13 stores the corresponding sound length, frequency, and sound. ^ Zhou (Chinese tones, that is, tones connected to Chinese syllables), pinyin symbols for each syllable, etc. The waveform overlapping accumulation unit 19 adds and accumulates CV syllables and vc (vv) syllabic segments. Synthetic speech output unit 20 Output synthetic speech η Application of the present invention. [Tai2 wanl shi4 yi2 ge5 mei3 li4 de5 6ao3 dao3] (Taiwan is a beautiful treasure island) twisted text ------------- ini ---- order ------- (Please read the meanings on the back and then fill out this page) 1 83 A7

I order to read it in advance

Fill in the note of A

This page Page ▲ Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs h ^ A7 --------- B7 _ V. Description of the Invention (8) Among the four Chinese words, the first Chinese syllable in Chinese. The boost ^ heavy accumulating unit 19 is based on the detailed information of the [^ syllable and vc (vv) consonant segment from the register 13, which includes the syllable length, syllable sub-sound length, syllable start position, syllable end position, The frequency value of the eight segments of the syllable, the type of syllable tone, the type of syllable consonant, the type of syllable vowel, the position of the syllable in the string π, the CV syllable number of the syllable, and the vc (vv) concatenated number of the syllable, etc. CV syllable and vc (vv) consonant waveform overlap and accumulate. Finally, the synthesized speech output unit 20 outputs the final synthesized speech. Chapter 7: Explains how to do the liaison processing for the ".. Taiwan" string according to the preferred embodiment of the present invention. First, the base material and mark data of each cv syllable and its liaison section are stored in the temporary storage.器 装置 13。 In the device unit 13. Then, calculate the frequency and length of the "Taiwan," and "Bay" and the consonant "Qiao Xong" syllables according to the prosody rules, so that the waveforms of these syllables and the consonant are overlapped and accumulated to generate the "Taiwan" character. The waveform of the string. Because the "Taiwan" character _ is synthesized based on the base period data, its sound length and frequency can be changed, and useful memory space can be saved. The example can solve the problem of liaison in the conventional technology. By detecting the liaison segment of the character string and superimposing and accumulating the syllable waveform of the character string and the syllable waveform before and after, it is possible to obtain a natural synthesizer. In addition, the pitch and frequency of a character string can also be changed, so that character strings with various tones and pitches can be generated, and useful memory space can be saved.

I —II --- *, -------— — — — — — — — < Please read the notes on the back before filling in this page) A7 451 1 83 ______ Β7 ____ 5. Description of the invention (9) Although the present invention has been thoroughly explained with the preferred embodiments and the drawings to which it is referred, it must be understood that various changes and modifications can be made in the technology as recognized by those skills. It must be understood that, as with the scope of the attached patent application, various changes and modifications can be included within the scope of the invention without departing from the principles of the invention. I I ------- ^ I I --- II -------- (Please read the precautions on the back before filling out this page) Printed by the Intellectual Property Bureau of the Ministry of Economic Affairs on consumer cooperation Du printed 12 This paper size is applicable to the national standard (CNS) A4 specification (2) 0 * 297 mm

Claims (1)

Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 6. Scope of patent application I. A speech processing device for Chinese speech synthesis, which includes: —s 可 典 S 自 忆 体 'to store a string of multiple characters and Dictionary of the corresponding Pinyin notation: A storage unit 'is used to store the basic weekly data of various Chinese syllables and connected syllables, the corresponding Chinese syllables and the corresponding phonetic syllables, and the various + wen syllables And the start and end positions of the consonants and vowels of these consonant segments; a character splitting unit, which analyzes a pinyin round-in sentence to be used for speech synthesis based on the dictionary stored in the dictionary memory and splits The sentence is a plurality of character strings; a syllable analysis unit is used to determine which character from the character analysis unit according to the storage unit _ must be processed in conjunction to detect the word that must be processed in conjunction A liaison segment of a metastring; and a speech synthesis unit for inserting the detected liaison segment between the syllables of the character string in the round-in pinyin sentence and generating a synthesized speech . 2. The continuous sound processing device of item 1 in the scope of patent application, wherein the storage unit contains 409 Chinese syllables with the first sound in Chinese. 3. If the liaison processing device of item i in the scope of patent application, wherein the liaison segments stored in the storage unit are the starting syllables of the subsyllables of some Chinese character strings " A liaison processing device, wherein the liaison segments stored in the storage unit are the starting syllables of the secondary syllables of some Chinese character strings. I-(Please read the notes on the back before filling this page) -------- Order --------- Line · 13