JPH10149188A - Text voice synthesizer - Google Patents

Abstract

PROBLEM TO BE SOLVED: To form almost natural voice synthesization concerning limited sentence examples. SOLUTION: Concerning a text voice synthesizer for regularly synthesizing arbitrary sentences in voice, this device is provided with a word dictionary part 62 storing a lot of words and having identification attributes in partial words so as to identify them from the other words, sentence example pattern table 66 having plural sentence example patterns composed of the string of identification attributes, and control part 65 for collating a word string provided from a language processing analytic part 63 with the sentence example pattern and controlling inserted voice synthesization or regular synthesization. A device for performing the inserted voice synthesization is provided with an insert table 73 with intonation composed of conjugation for conjugating plural words and the intonations of inserted word strings and an inserted rhythm generating part 74 for generating a pitch pattern while using the intonation of inserted word string and connecting a waveform for the unit of a voice according to this pitch pattern.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a text-to-speech synthesizing apparatus, and more particularly to a method for forming a speech synthesis that is almost natural for a limited sentence example.

[0002]

2. Description of the Related Art In FM data multiplex broadcasting, character information data is multiplexed in a broadcast. In recent years, VICS (road traffic information communication system) which multiplexes traffic congestion information data into a car navigation system for a vehicle. ) Service has begun. The VICS traffic congestion information includes character information along with map information. On the receiver side, this traffic congestion information is visually displayed.

Furthermore, it is preferable from the viewpoint of driving of a vehicle that not only the character information of the traffic congestion information be displayed visually, but also that it be synthesized by voice and appeal to hearing. Since the character information of the traffic congestion information is a sentence mixed with kanji and kana, it can be converted to voice by a text-to-speech synthesizer. In this text-to-speech synthesis apparatus, a sentence mixed with kanji kana is analyzed and processed into a phonetic character string composed of phoneme symbols and prosodic symbols, and the duration of each phoneme with respect to the phonetic character string, the pitch in which accent is superimposed on intonation A prosody such as a pattern is generated. Phoneme waveforms are connected based on the rules of this prosody to synthesize speech.

[0004]

However, since the connection rules between the above phoneme units are complicated, the pitch pattern obtained deviates from the natural pitch pattern,
There is a problem that it is difficult for voice to be natural voice. If the synthesized voice of the traffic information is an unnatural voice, even if the traffic information is visual to audible information, its usefulness cannot be exhibited.

SUMMARY OF THE INVENTION Accordingly, an object of the present invention is to provide a speech synthesizer capable of making a synthesized speech of character information in traffic information close to nature.

[0006]

In order to solve the above-mentioned problems, the present invention provides a text-to-speech synthesizing apparatus for synthesizing an arbitrary sentence into speech in a rule-by-speech manner. A language processing analysis unit for converting, and storing a large number of words for conversion into the phonetic character string, and identifying a part of the word to obtain the word identification attribute string with an identification attribute for identifying the word from other words. A sentence example pattern table having a sentence example pattern including the word dictionary unit and the identification attribute sequence, and determining whether the word identification attribute sequence obtained from the language processing analysis unit matches the sentence example pattern in the sentence example pattern table. If they match, embedded speech synthesis is performed. If they do not match, a pitch pattern is generated based on the intonation of the phonetic character string obtained by the language processing analysis unit. And a control unit for controlling the rule synthesis using the switch pattern. The apparatus for performing the inlay speech synthesis includes a plurality of corresponding words included in the column of the identification attributes, and forms one sentence. An inset sentence example table having a plurality of inset sentence example patterns composed of a plurality of connective words that connect a plurality of words and a sentence-specific intonation, and a pitch pattern is generated based on intonations held in the inset sentence example table. It is characterized by having an inset type prosody generation unit for connecting a waveform of a speech unit constituting a phonogram string of a word string inserted using a pitch pattern. By this means, when the sentence example of the traffic character information is limited as in VICS, it is easy to obtain the intonation of the announcer for the limited sentence example. For this reason, the voice of the voice character information of a more natural intonation can be obtained than the language processing analysis.

[0007]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a diagram showing a text-to-speech synthesis apparatus according to an embodiment of the present invention. An FM receiver mounted on a vehicle, as shown in the figure, a receiver 2 connected to an antenna 1, a switch 3 and a speaker 4.
, And FM broadcast is output. The decoder 2A of the receiver 2 converts character data such as news, VICS
Character data and map data are separated. The display 5 displays the map data and the character data separated by the decoder 2A. The text-to-speech synthesizing section 6 receives the character data and synthesizes it with a conventional text-to-speech synthesizing method or a voice-to-speech synthesizing method, and the synthesized voice is output via the switch 3 and the speaker 4.

FIGS. 2 and 3 are diagrams showing the configuration of the text-to-speech synthesizing unit 6 shown in FIG. As shown in FIG. 2, the character data normalization unit 61 inputs a character mixed with kanji kana, converts the character into normalized character data, and outputs this to the language processing analysis unit 63. The conversion table 62 is a conversion table for converting a sentence mixed with kanji and kana into normalized character data, and is referred to by the character data normalization unit 61. The language processing analysis unit 63 inputs the normalized character data, converts it into a series of words, identifies parts of speech and inflected forms, and sets prosody information such as pause setting, reading addition, accent setting, etc.
The attribute of a specific word, which will be described later, is analyzed, and the attribute and prosodic (kana) phonetic characters are output. The word dictionary unit 64 has tens of thousands to hundreds of thousands of words and describes grammatical information, readings, accent forms, and the like. The specific words include, for example, A, B, C, D, and the like. The type code is given as follows, and is referred to by the language processing analysis unit 63.

The control unit 65 uses the sentence example pattern table 66 to convert the attributes and the prosody-signed (kana) characters output from the language processing analysis unit 63 into a phonogram string for rule synthesis or a phonogram string in the embedded format. It is determined whether there is. If it is determined that the phonetic character string for rule synthesis is used, control of normal text-to-speech synthesis is performed as follows. That is, the rule synthesis type prosody generation unit 67 generates intonation from the position of the pause in the phonogram character string for rule synthesis, and generates a pitch pattern by overlapping the intonation with the accent. The prosody table 68 describes rules for setting poses, accents, etc., and is referred to by the rule synthesis type prosody generation unit 67. The phoneme duration generation unit 69 sets the duration of the phonemes constituting the phonogram to realize natural timing (rhythm) of the utterance. The phoneme length table 70 describes a unique time length rule of the phoneme in consideration of the influence of the adjacent phoneme, and is referred to by the phoneme duration generation unit 69. When connecting the phonemes constituting the rule-synthesizing phonogram string to synthesize a continuous speech, the waveform connection unit 71 performs an interpolation process in which characteristic parameters such as the phoneme duration and the pitch pattern are expressed between phonemes. The continuous sound that has been subjected to the interpolation processing is output as PCM data. The phoneme dictionary unit 72 stores the phoneme waveform data as a synthesis unit and is referred to by the waveform connection unit 71.

Next, the description is returned to the word dictionary section 64. An example of the character information of the traffic information has a limited sentence pattern as follows. The sentence has been simplified to appeal to the visual sense. "The westbound of the second Shinmei road is near Tazu, traffic congestion is 2km." According to the normalized character data to the language processing analysis unit 63, the "second Shinmei road" is replaced with "Dainishin Meadoro" and Kana Converted, means road name, "westbound" is converted to "Nishiyuki" and kana, and means direction, "Tamazu" is converted to "matsu" and kana, means place name, "congestion"
Is converted to "jutai" and kana, meaning degree,
“2 km” is converted to “niklo” and kana, which means a distance.

[0011] In this case, the word dictionary unit 64 stores these dainisin medoro, Nishiyuki, Tatsuma, Jutai,
Have words such as ni-kilo, but these words include, for example,
A code having an attribute "A" is given to a word meaning a road name. Similarly, a word meaning direction is “B”, a word meaning place name is “C”, a word meaning degree is “D”,
A code having an attribute “E” is given to a word meaning a distance. Other attributes include "F" for the word meaning cause, "G" for the word meaning incident, "H" for the word meaning date, "I" for the word meaning time,
"J" for the word meaning time, "K" for the word meaning regulation, "L" for the word meaning speed, "M" for the word meaning attention, and "N" for the word meaning transportation means "O" is assigned to a word meaning an entrance, "P" is assigned to a word meaning an area, "Q" is assigned to a word meaning an alarm, and the like.

In the sentence example pattern table, word strings of limited sentence examples are created and held as follows. Word string Attribute string Sentence example number Route name + direction + place name + degree + distance A + B + C + D + E 1 Route name + place name + direction + degree + distance A + C + B + D + E 1 Route name + place name + direction + distance + degree A + C + B + E + D 1 Route name + direction + place name + distance + Degree A + B + C + E + D 1 Route name + direction + place name + place name + degree + distance A + B + C + C + D + E 2 Route name + direction + place name + place name + distance + degree A + B + C + C + E + D 2 Route name + direction + cause + place name + distance + degree A + B + F + C + E path + Cause + Place name + Degree + Distance A + B + F + C + D + E 3... The control unit 63 checks the sentence example pattern table 66 when one sentence composed of the attribute and the Kana character with the prosody symbol is input. Example pattern table 6
In the case where the sentence example does not completely match the sentence example, the rule synthesis type prosody generation processing is selected for the following pitch generation processing. On the other hand, if it matches one of the sentence examples, the inset type prosody generation processing is selected and a phonetic character string in the inset format is generated.

An inset format phonetic character string is formed from a sentence number and the reading of an inset word. Although the sentence example pattern table 66 has four examples for the above sentence example number 1, the sentence example table may have only one example because the information content is the same only by changing the word position. This is to reduce the memory capacity.

The inset sentence example table 73 shown in FIG. 3 is formed of phonetic character strings of connecting words, such as particles, auxiliary verbs, and verbs, connecting the words constituting the sentence examples, corresponding to each of the sentence examples in the sentence example pattern table 64. For example, in the example at the top of the sentence example number 1, it is composed of "No", "C", "Fukinde" and "Death". In this case, the inset type prosody generation unit 74 converts each word of the attribute string of the word constituting the inset format phonetic character string between the phonetic character strings of connective words such as particles, auxiliary verbs and verbs in the sentence example parameter table. Fit it. In the case of the above sentence example 1, an inset phonetic character string such as “A“ NO ”B“ H ”“ C ”Fukinde“ DE “Death.” ”Is obtained.

Since the character information of the traffic information has a visual structure, the word is completed as a sentence inserted between phonetic character strings of connecting words such as particles, auxiliary verbs, verbs, etc., and appeals from sight to hearing. Change to Next, in the inset sentence example table 73, an appropriate word is inserted between connecting words such as particles, auxiliary verbs, and verbs of each sentence example to form sentences, and each sentence is uttered by an announcer, and intonation is analyzed.
This is kept for each sentence. The inlaid type prosody generation unit 74 refers to the prosody table 75 having the same configuration as the prosody table 68 for the accent and pause of the inlaid word, and refers to the inset sentence example table 73 for the intonation of the entire sentence to superimpose the intonation and accent. To form a pitch pattern. In the same manner as described above, the waveform generation unit 71 sets the phoneme duration for the phonemes of the inlaid phonetic character string, sets the pitch pattern from the inlaid prosody generation unit 74, and connects the waveforms of the phonemes.

Therefore, according to the present invention, since a pitch pattern is set for a limited example of traffic information, natural speech synthesis can be obtained. In addition, news other than traffic information, for example, "Major private railways also divide 10,000 yen, wage increase 9950 yen, 3.38
It is difficult to assume a sentence example in advance for information such as "%", and speech synthesis is performed using conventional rule synthesis instead of inset synthesis.

In the configurations shown in FIGS. 2 and 3, the depolarizer buffer, the input buffer, and the output buffer may be shared between the rule combination and the fitting combination. This is because two types of algorithms can be used without increasing the data memory. Further, the sentence example pattern table 66 and the sentence example parameter table 73 may be separated from the speech synthesis program and installed in a flash memory or the like. This is to make it easy to upgrade the base of the sentence example data.

[Brief description of the drawings]

FIG. 1 is a diagram showing a text-to-speech synthesis apparatus according to an embodiment of the present invention.

FIG. 2 is a diagram showing a configuration of a text-to-speech synthesizing unit 6 shown in FIG.

FIG. 3 is a diagram showing a configuration of a text-to-speech synthesis 6 shown in FIG.

[Explanation of symbols]

 6: Text-to-Speech Synthesizing Unit with Inset Type 61: Normalization Unit for Character Data 62: Conversion Table 63: Language Processing Analysis Unit 64: Word Dictionary Unit 65: Control Unit 66: Sentence Example Pattern Table 67: Rule Synthesizing Prosody Generation Unit 68 , 75: Prosody table 69: Phoneme duration generation unit 70: Phoneme length table 71: Waveform connection unit 72: Phoneme dictionary unit 73: Inset table with intonation 74: Inset type prosody generation unit

Claims (8)

[Claims] 1. A text-to-speech synthesizer for regularly synthesizing an arbitrary sentence into speech, comprising: a language processing analysis unit for converting the sentence into a phonetic character string and a word identification attribute string; Memorize many words in
A word dictionary unit in which a part of a word has an identification attribute for identifying the word from other words to obtain the word identification attribute sequence; a sentence example pattern table having a sentence example pattern including the identification attribute sequence; It is determined whether or not the word identification attribute sequence obtained from the language processing analysis unit matches the sentence example pattern in the sentence example pattern table, and if it matches, the embedded speech synthesis is performed. A control unit that generates a pitch pattern based on the intonation of the obtained phonogram character string, and performs control so that the rule synthesis is performed using the pitch pattern. A plurality of connecting words that connect a plurality of words so as to form one sentence by inserting a plurality of corresponding words included in a column of identification attributes and a sentence-specific intonation An inset sentence example table having a plurality of inset sentence example patterns, and a pitch pattern is generated based on intonations held in the inset sentence example table, and a phonetic character string of the inset word string is formed using the pitch pattern. A text-to-speech synthesizer, comprising: an inset-type prosody generation unit for connecting waveforms of speech units to be synthesized. 2. The word dictionary unit according to claim 1, wherein a part of the word has an identification attribute for identifying the word from other words.
Road name, direction, place name, degree, distance, cause, incident, due date, time, time, regulation, speed, caution, transportation,
2. The text-to-speech synthesis apparatus according to claim 1, wherein words that mean an entrance, an area, a warning, and the like are set. 3. The text-to-speech synthesis apparatus according to claim 1, wherein the inlaid sentence example table forms intonation information unique to the sentence example by an intonation generated by an announcer. 4. The text-to-speech synthesis apparatus according to claim 1, wherein the control unit refers to the sentence example pattern table and outputs a corresponding unique sentence example number. 5. The text voice according to claim 1, wherein the sentence example pattern table sets any one of a plurality of similar word strings in which the words constituting the word string are the same and the order is different. Synthesizer. 6. The text-to-speech synthesis apparatus according to claim 1, wherein the rule synthesis and the inlay speech synthesis use a common temporary buffer, input and output buffers. 7. The text-to-speech synthesizing apparatus according to claim 1, wherein the sentence example pattern table and the inlay table with intonation are formed by a flash memory. 8. The text-to-speech synthesis apparatus according to claim 1, wherein the word dictionary unit indicates, as a code, an identification attribute for identifying a part of the word from other words.