JPH0679235B2 - Speech synthesizer - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a speech synthesizer for synthesizing speech, and in particular, it is based on a rule in a set of phoneme connection rules that defines how to connect two phonemes. The present invention relates to a speech synthesizing device having a rule synthesizing unit that connects two adjacent phonemes in a phoneme sequence to generate a speech synthesizing parameter.

(Prior Art) FIG. 4 shows a configuration of an example of a conventional speech synthesizer of the type described above. This speech synthesizer includes an input unit 1 into which an arbitrary character string or a musical notation string (hereinafter abbreviated as "character string or the like") is input, and rule synthesis that generates a speech synthesis parameter for the input character string or the like. Unit 102 and synthesizer 3 for outputting a rule-synthesized voice based on the voice synthesis parameter
It consists of and.

The rule synthesizer 102 will be described.

The character string and the like input from the input unit 1 is syntactically analyzed by the character string analysis unit 21. In this syntactic analysis, the words in the character string or the like are identified with reference to the word dictionary 22. Thereby, the phoneme symbol string corresponding to the character string or the like is determined, and the accent information is given to the phoneme in the phoneme symbol string corresponding to the accent position of the word in the character string or the like. Further parsing determines the intonation of the sentence or the entire sentence. The phoneme symbol string is input to the speech synthesis parameter generation unit 123 and processed for speech synthesis.

The processing in the voice synthesis parameter generation unit 123 will be described with reference to the flowchart of FIG. The speech synthesis parameter generation unit 123 first inputs the phoneme symbol string generated by the character string analysis unit 21 (step). A phoneme feature parameter that characterizes a phoneme corresponding to each phoneme of the input phoneme symbol string is extracted from the feature parameter file 24 (step). A phoneme connection rule for connecting two adjacent phonemes in the phoneme symbol string is extracted from the rule file 125 (step). The rule includes control structure and control information for connecting phoneme feature parameters of two phonemes. By applying each of the rules, a phoneme connection feature parameter that connects two adjacent phonemes is generated (step). A voice synthesis parameter is generated from the phoneme feature parameter and the phoneme connection feature parameter thus obtained, and the information on the accent and intonation previously obtained by the character string analysis unit 21 (step).

(Problems to be Solved by the Invention) In the conventional speech synthesizer described above, a phoneme connection rule is created for each of two pairs of adjacent phonemes that can occur and stored in the rule file 125. There was a need. That is, the phoneme connection rules are roughly classified into (i) phoneme connection rules for connecting vowels to vowels, (ii) phoneme connection rules for connecting vowels to consonants, and (iii) consonant connections to vowels. There are three types of phoneme connection rules for doing (that is, connecting syllables).

In order to hold many such rules, it was necessary to give the rule file a very large storage capacity. Therefore, there is a drawback that the space and cost occupied by the rule file become large.

The present invention has been made in view of the above problems, and an object thereof is to provide a speech synthesizer capable of storing and retaining a phoneme connection rule in a small space and at a small cost.

(Means for Solving Problems) A speech synthesis apparatus according to the present invention uses two adjacent phonemes in a phoneme sequence based on a rule in a set of phoneme connection rules that defines how to connect two phonemes. 1. A speech synthesis apparatus having a rule synthesizing unit for connecting phonemes to generate a speech synthesis parameter, wherein the rule synthesizing unit is a phoneme (first phoneme) in a phoneme sequence and another phoneme immediately after the certain phoneme. Means for determining whether or not a phoneme connection rule for connecting (second phoneme) exists in the set, and a phoneme for connecting the second phoneme to the first phoneme by the determining means When it is determined that the connection rule does not exist, the phoneme connection rule for connecting the first phoneme to the second phoneme existing in the set is applied in a temporally reverse direction to apply the phoneme connection rule to the first phoneme. And a means for connecting the second phoneme to the phoneme, whereby the above object It is achieved.

Figure 3 shows the voice waveform of male natural voice / ama / and the formant locus obtained by analyzing the waveform. The analysis method is as follows. The sampling frequency of the audio signal was 10 KHz, a Hamming window was applied for one frame of 20 ms, and pre-emphasis with a coefficient of 0.98 was applied. A linear prediction analysis is performed by the 12th-order autocorrelation method, a solution of a higher-order equation whose coefficient is the linear prediction coefficient obtained by the analysis is obtained, and the solution is used to estimate the formant trajectory of the McCandless formant trajectory (“ An Algorithmfor Automatic
Formant Extraction Using Linear Prdiction Spectr
a ", IEEE ASSP, Vol.22, No.2, April, 1974).

The formant transition from the vowel / a / to the consonant / m / nasalized part and the transition from the consonant / m / nasalized formant to the vowel / a / formant are slightly different, but generally Can be regarded as symmetric in. The same phenomenon is observed in the combination of consonant / m / with other vowels and the combination of consonant / s, n, h, r, w / with vowels. Therefore, for these combinations, only the phoneme connection rule that connects vowels to consonants is prepared, and the consonant connection to vowels applies the phoneme connection rule that connects vowels to consonants in the opposite time direction. However, the quality of the synthesized voice does not deteriorate so much. Partial autocorrelation (PARCOR) coefficient, linear prediction coefficient, line spectrum pair (LS
The same can be said when the P) parameter and vocal tract cross-sectional area ratio are used. When synthesizing the above-mentioned speech / ama / using formant as a feature parameter, the transition from / a / to / m / to the nasalized portion is performed from the / m / nasalized portion to / a /
When the transition to was inverted and used in the time direction, a fairly clear and natural synthesized speech was obtained.

(Example) Hereinafter, the present invention will be described with reference to Examples.

FIG. 1 shows a block diagram of an embodiment of the speech synthesizer of the present invention. The same parts as those of the conventional speech synthesizer shown in FIG. 4 are designated by the same reference numerals. The speech synthesizer of this embodiment basically has a configuration similar to that of the conventional speech synthesizer shown in FIG. 4, and has an input unit 1 for inputting a character string or the like, and speech synthesis for the character string or the like. Rule synthesizer that generates parameters
2, and a synthesizing section 3 for outputting a rule-synthesized speech based on the speech synthesis parameter.

The character string analyzing unit 21 of the rule synthesizing unit 2 parses a character string or the like sent from the input unit 1 to convert it into a phoneme symbol string, and further extracts accent and intonation information. The word dictionary 22 is referred to when parsing. The phoneme symbol string, accent, and intonation information are sent to the voice synthesis parameter generation unit 23.

The voice synthesis parameter generation unit 23 extracts a phoneme feature parameter that characterizes each phoneme of the phoneme symbol string from the character string analysis unit 21 from the feature parameter file 24, a feature parameter extraction unit 230, and two adjacent phoneme symbol strings in the phoneme symbol string. Phoneme connecting means 231 for generating a phoneme connection characteristic parameter connecting the two phonemes by applying the corresponding phoneme connection rule in the rule file 25 to the phonemes of the phoneme, and a speech synthesis parameter from those characteristic parameters. A synthetic parameter generation means 235 is provided. The phoneme connection means 231 determines whether or not the phoneme connection rule corresponding to a pair of two phonemes to be connected (referred to as "phoneme A-phoneme B") is included in the rule file 25. 232 and determination means 232
If the phoneme connection rule is determined to be “included”, the phoneme connection rule is extracted from the rule file 25, and the phoneme connection feature parameter for applying the rule to connect the phoneme B to the phoneme A is generated. 233 and the phoneme connection rule for connecting the phoneme A to the phoneme B when the judgment means 232 judges that the phoneme A is not included, from the rule file 25,
Reverse direction phoneme connecting means 234 for connecting the phoneme B to the phoneme A by applying the rule in the reverse direction in time.

Formant frequencies and formant bandwidths are used as the phoneme feature parameters stored in the feature parameter file 24, but instead of partial autocorrelation (PARC
OR) coefficient, linear prediction coefficient, line spectrum pair (LSP) parameter, vocal tract cross-sectional area ratio, etc. can also be used.

The rule file 25 holds a phoneme connection rule for connecting two adjacent phonemes in a phoneme symbol string.
The rules for all phoneme pairs are not retained. Phoneme connection rules are prepared for all vowel-vowel and consonant-vowel phoneme pairs, while for vowel-consonant phoneme pairs the consonant prefix phoneme is / s,
The phoneme connection rules for n, h, m, r, w / are not prepared.
In the case of unvoiced plosives with consonants of / p, t, k / in a vowel-consonant pair, there is no sound immediately before the start of a consonant, so the rule of vowel-silence is used regardless of the type of consonant. ing. As a result, the storage capacity of the rule file 25 is considerably smaller than the storage capacity of the rule file 125 of the conventional speech synthesizer.

The processing in the speech synthesis parameter generation unit 23 will be described based on the flowchart of FIG. If the correspondence with FIG. 1 is shown, the steps ~ represent the processing by the phoneme connecting means 231, of which the steps ~ are the judgment means 232,
Steps 1 to 3 represent processing by the reverse phoneme connecting means 234, and steps represent processing by the forward phoneme connecting means 233.

(1) The phoneme symbol string generated by the character string analysis unit 21 is read (step), and the phoneme characteristic parameter corresponding to each phoneme in the phoneme symbol string is taken out from the characteristic parameter file 24 (step).

(2) A pair of adjacent two phonemes is cut out from one end of the phoneme symbol string (step). For the sake of explanation, this set of phonemes will be referred to as phoneme A-phoneme B.

In steps (3) to, it is determined whether the phoneme connection rule corresponding to the phoneme A and the phoneme B exists in the rule file 25.
Phoneme A-phoneme B is a vowel-consonant (in the step "Ye
s ”), if there is no phoneme connection rule corresponding to the phoneme set in the rule file (“ No ”in step), go to step. If phoneme A-phoneme B is not a vowel-consonant ("No" in step), or if it is a vowel-consonant, there is a corresponding phoneme connection rule ("Ye in step"
s ”), go to the step.

(4) If there is no phoneme connection rule corresponding to phoneme A-phoneme B, phoneme A-phoneme B is inverted to phoneme B-phoneme A (step), and phoneme connection rule corresponding to phoneme B-phoneme A is set. A phoneme connection feature parameter for connecting the phoneme A to the phoneme B is generated based on the rule file 25 (step), the generated phoneme connection parameters are rearranged in the reverse direction in the time direction, and the phoneme A is phoneme B. Is used as a phoneme connection feature parameter (step).

(5) If a phoneme connection rule corresponding to phoneme A-phoneme B is present in the rule file, that rule is taken out from the rule file 25, and a phoneme connection feature parameter for connecting phoneme B to phoneme A is generated based on this rule. Do (step).

(6) If all adjacent pairs of two phonemes in the phoneme symbol string are not completely connected (“No” in step), the next pair of adjacent two phonemes is cut out (step) and the process returns to the step. If all adjacent two phonemes are connected (“Yes” in step), the phoneme feature parameters and phoneme connection feature parameters obtained above and the character analysis unit
A voice synthesis parameter is generated using the accent and intonation information obtained in step 21 (step).

In the above-described embodiment, the phoneme connection rule for connecting consonants to vowels is prepared in the rule file, and the phoneme connection rule for connecting consonants to vowels is omitted. It is also possible to adopt another configuration such as a configuration in which a phoneme connection rule for connection is prepared and the rule for connecting a vowel to a consonant is omitted.

(Effects of the Invention) According to the present invention, the capacity of the memory for holding the phoneme connection rule can be kept small without deteriorating the quality of the synthesized speech, thus reducing the space and cost for the memory. A speech synthesizer is provided.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of an embodiment of the speech synthesis apparatus of the present invention, FIG. 2 is a flow chart showing an example of the processing procedure in the speech synthesis parameter generation unit of the speech synthesis apparatus of the present invention, and FIG. FIG. 4 is a block diagram showing the configuration of an example of a conventional speech synthesizer, and FIG. 5 is a speech synthesis of the conventional speech synthesizer. It is a flow chart which shows the example of a processing procedure in a parameter generation part. 1 ... Input unit, 2 ... Rule synthesis unit, 3 ... Synthesis unit, 21 ... Character string analysis unit, 22 ... Word dictionary, 23 ... Speech synthesis parameter generation unit, 24 ... Feature parameter file, 25 ...... Rule file, 230 ...... Characteristic parameter extraction means, 231 ...... Phoneme connection means, 232 ...... Judgment means, 233 ...... Forward phoneme connection means, 234 ...... Reverse phoneme connection means, 235 …… Synthesis parameter generation means.

Claims (1)

[Claims] 1. A rule synthesizing unit for connecting two adjacent phonemes in a phoneme sequence to generate a speech synthesis parameter based on a rule in a set of phoneme connection rules defining how to connect two phonemes. A phoneme connection for connecting a certain phoneme (first phoneme) in a phoneme sequence to another phoneme (second phoneme) immediately after the certain phoneme. Means for determining whether a rule exists in the set, and if the determining means determines that there is no phoneme connection rule for connecting the second phoneme to the first phoneme Means for connecting the second phoneme to the first phoneme by applying a phoneme connection rule for connecting the first phoneme to the second phoneme existing in Speech synthesizer.