JPH01211799A - Regular synthesizing device for multilingual voice - Google Patents

Abstract

PURPOSE:To obtain a voice signal which has individual features as to plural languages by providing a regular synthesizing means which synthesizes the voice signal of a 1st standard speaker with respect to the plural languages and a voice converting means which converts a voice signal outputted by a selecting means into the voice signal of a 2nd speaker to whom individual features are to be added. CONSTITUTION:A switching part 100 selects one of languages L1-Ln, e.g., L1 and outputs the voice signal s11 of a standard speaker A1 of the selected language L1 from a multilingual regular synthesizing group 104 to a voice conversion part 101. The voice quality conversion part 101 receives the voice signal s11, refers to data on the speaker B whose voice is to be given individual ity in a voice individual information file 102, and converts the voice signal s11 of the standard speaker A1 into the voice signal s4 of the speaker B, which is outputted. Consequently, the individual features of speaker are given to regu larly synthesized voices of respective languages.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a speech rule synthesis device, and particularly to a rule synthesis device that handles multiple languages.

[Prior Art and Problems to be Solved by the Invention] Conventionally, rule synthesis systems have been developed for each language (eg, Japanese, English, German, etc.). However, these rule synthesis systems are designed to be used only by native speakers.
It does not have functions that non-native speakers can use.

In other words, the rule synthesis system developed for each language is
Since the units of speech used for synthesis and prosodic control rules are created based on those of native speakers, the synthesized speech using these rules can output more fluent speech than the speech spoken by non-native speakers. The voice qualities that can be output by these rule synthesis systems are limited to one or several types (male voice, female voice, childlike voice, elderly voice, etc.).

Therefore, it is not possible for a rule synthesis system to synthesize speech on behalf of a certain person, that is, to synthesize speech that is close to the voice quality of a certain person's voice and output speech as if that person were speaking in that language. There was a problem.

The reasons for this are: (1) The speech units used for rule synthesis are created from hundreds of speeches produced by a certain speaker, which places a heavy burden on the speaker.

(2) Creation of speech units is difficult and labor-intensive to perform fully automatically, and it is virtually impossible to create speech units for a large number of speakers.

(3) If a unit set of speech used for rule synthesis is created for each speaker, the amount of memory to be stored will become enormous. − and so on.

This invention has been made to solve the above-mentioned problems, and aims to give the speaker's personal characteristics to the rule-synthesized speech synthesized for each language.

[Means for Solving the Problems] A speech rule synthesis device that handles multiple languages according to the present invention receives character information signals from the outside for each of a plurality of languages, and generates a standard first speaker for each language. A plurality of rule synthesis means for synthesizing the speech signals of the first speaker of each language by referring to the speech unit signal set, and selecting one of the speech signals of the first speaker of each language. a selection means for outputting, a conversion signal file means storing conversion signals necessary for converting a first speaker's voice into a second speaker's voice to which it is desired to add personal characteristics to the voice, and a selection means. and voice conversion means for converting the first speaker's voice signal outputted from the first speaker into a second speaker's voice signal based on the conversion signal stored in the conversion signal file means.

[Operation] The speech rule synthesis device that handles multiple languages in this invention has the following features:
After the rule-synthesizing means rules-synthesizes the first speaker's speech signal that is the standard for that language, the speech conversion means converts the language selected by the selection means into a converted signal stored in the converted signal file means. Since the first voice signal is converted into a second speaker's voice signal having personal characteristics, it is possible to obtain a voice signal having personal characteristics for multiple languages.

[Embodiment of the Invention] FIG. 1 is a block diagram showing an embodiment of a rule synthesis device that handles multiple languages according to the present invention.

Referring to FIG. 1, this rule synthesis device includes an input section ]0 to no, a multilingual rule synthesis group 104, and a switching section 100
, voice quality conversion unit 101 , and voice personal information file 102
and an output section 103.

This rule synthesis device can handle N languages L1 to LN (eg, English, German, Chinese, Japanese, etc.).

The multilingual rule synthesis group 104 includes rule synthesis units that process signals from input units to which character information signals are given for each language. For example, for a certain language L1, a character information signal slO including character strings such as characters and accent types related to the language L1 and a prosodic signal is provided to the input unit 10 from the outside. A rule synthesis unit 1 including a unit set information file 12 of standard speaker A1's speech of language L1
1 receives the character information signal slO, refers to the voice unit set information file 12, and determines the standard speaker A1 of the language.
The audio signal sll of is synthesized according to the rules.

On the other hand, for languages other than the language L1, the speech signals s21 to snl of the standard speakers A2 to An of each language can be similarly synthesized by the rule synthesis units 21 to nl.

Here, audio unit set information files 12 to n2
are the standard speakers A1 to An of each language L1 to Ln.
Each of these is a database in which information on units of speech such as phonemes and syllables is stored in advance.

The switching unit 100 selects one of the languages L1 to Ln, for example, the language L1, and selects the speech of the standard speaker A1 of the selected language L1 output from the multilingual rule synthesis group 104. The signal S11 is output to the voice quality conversion section 101.

The voice quality conversion unit 101 receives the audio signal sll, refers to the data of the speaker B whose voice is desired to be personalized from the audio personal information file 102 registered in advance, and converts the audio signal of the standard speaker A1 into a voice signal of the standard speaker A1. sll is converted into speaker B's audio signal s4 and output.

As a voice quality conversion method in the voice quality conversion unit 101, a voice quality conversion method using vector quantization is used. This method converts the voice quality between a standard speaker, which is the standard for the rule synthesis unit, and a speaker who wants to add individuality to the voice, using a conversion codebook that is a correspondence between the codebooks of each speaker. It is something.

The conversion codebook includes the power, pitch frequency, and spectrum information of the speaker's voice to which individuality is to be added, and the characteristics of the voice are expressed discretely. The voice personal information file 102 in FIG. 1 includes the contents of this conversion codebook for each speaker whose voice is desired to be personalized.

In the following, - as an example, the language L1 is English, and a certain American speaker A1's speech signal s'll synthesized by the English rule synthesis unit 11 is converted into a Japanese speaker B's speech signal S4 to which individuality is to be added. The case of conversion will be explained.

FIG. 2 is a flow diagram showing the procedure for creating a conversion codebook.

Referring to Figure 2, below is the conversion code book 41.42
．． The procedure for finding 43 will be explained.

First, in steps 301 and 302, American speaker A1 and Japanese speaker I are made to utter the same word, and their respective voices are subjected to LPG analysis to determine power, pitch frequency, and spectral parameters. Next, in steps 303 and 304, the spectral parameters are vector quantized, and in steps 305 and 306
scalar quantize the power in steps 307 and 3
08, the pitch frequency is scalar quantized.

Step 309 uses the spectral parameters to take the time correspondence of the voices uttered by speaker A and speaker B.
DP matching is performed using the Double 5plit method. Based on the time-related information obtained here, in steps 310, 311 and 312,
The correspondence between speaker A and speaker B is determined for each feature, and a histogram is created. The spectral parameter and power conversion codebook 4L 43 is obtained by linear combination of the feature vectors of speaker B using this histogram as weight. Further, the pitch frequency conversion codebook 42 is created using the feature vector of speaker B that gives the maximum value of this histogram.

FIG. 3 is a flow diagram showing the voice quality conversion procedure in the voice quality conversion section 101.

Referring to FIG. 3, a voice quality conversion method using a conversion codebook will be described below using language L1, that is, English, as an example. The speech sll of speaker A1 is subjected to LPG analysis in step 401 to extract power, pitch frequency and spectral parameters. Next, step 4
The spectral parameters from speaker A1's spectral codebook are vector quantized in step 02, the powers from speaker A1's power codebook are scalar quantized in step 404, and the powers from speaker A1's pitch frequency codebook are vector quantized in step 404. The pitch frequency of is scalar quantized. In the process of decoding these quantized parameters, the aforementioned transformation codebooks 41.42.43 are used. That is, in step 405, the spectrum conversion codebook 41 from speaker A1 to speaker B is used, and in step 406,
The power conversion codebook 43 is used, and the pitch frequency conversion codebook 42 is used in step 407. Then, using each converted parameter, step 408
The voice of speaker B is synthesized. As a result, a regular synthesized English sound having the voice quality of Japanese speaker B is obtained.

FIG. 4 is a schematic block diagram showing the hardware configuration of a rule synthesis system including a rule synthesis device that handles multiple languages according to the present invention.

Referring to FIG. 4, this rule synthesis system consists of amplifier 1
, a low-pass filter 2 , an A/D converter 3 , and a computer system 4 . The amplifier 1 is for amplifying an input audio signal, and the low-pass filter 2 is for removing aliasing noise from the amplified audio signal. A/D
The converter 3 converts the audio signal into a 16-bit digital signal using a 12 kHz sampling signal. The computer system 4 includes a rule synthesis device (arithmetic processing unit) 5, a magnetic disk 6, terminals 7, and a printer 8. The speech rule synthesis device that handles multiple languages according to the present invention is configured within the rule synthesis device 5 shown in FIG.

The above-described rule synthesis device that handles multiple languages is particularly useful as a rule synthesis system for multilingual voice proxy, which outputs speech in place of a speaker's utterance of multiple languages.

[Effects of the Invention] As described above, according to the present invention, the rule synthesis means for synthesizing the speech signal of the first speaker, which is standard for multiple languages, and the speech signal output from the selection means are adapted to personal characteristics. and a voice converting means for converting the voice signal into the voice signal of the second speaker to whom it is desired to impart a voice signal, it is possible to obtain a voice signal having personal characteristics in multiple languages.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of a rule synthesis device that handles multiple languages according to the present invention. FIG. 2 is a flow diagram showing the procedure for creating a conversion codebook. Figure 3 shows
FIG. 3 is a flow diagram showing a voice quality conversion procedure in a voice quality conversion unit. FIG. 4 is a schematic block diagram showing the hardware configuration of a rule synthesis system including a rule synthesis device that handles multiple languages according to the present invention. In the figure, 1 is an amplifier, 2 is a low-pass filter, 3 is an A/D converter, 4 is a computer system, 5 is a rule synthesis device, 10 to no are input sections, 11 to nl are rule synthesis sections, 12 to n2 are Voice unit set information file, 100 is a switching unit, 101 is a voice quality conversion unit, 102 is a voice personal information file, 103 is an output unit, slO or s
nO is a text information signal, sll to snl are voice signals of speakers A1 to An, s3 is a voice personal information signal of speaker B, and s4 is a voice signal of speaker B.

Claims (1)

[Scope of Claims] For each of a plurality of languages, the system includes a set of standard voice unit signals of a first speaker, receives a character information signal from the outside, and refers to the set of voice unit signals. a plurality of rule synthesis means for synthesizing speech signals of a first speaker; and a plurality of rule synthesis means connected to the plurality of rule synthesis means, for selecting one language from the plurality of languages, and for selecting the first language for the selected language. a selection means for outputting a voice signal of a second speaker; and a conversion signal stored therein necessary for converting the voice of the first speaker into the voice of a second speaker whose voice is desired to have personal characteristics. a converted signal file means for receiving the synthesized voice signal of the first speaker in the selected language, and converting the voice of the first speaker based on the converted signal stored in the converted signal file means; a voice conversion means for converting a signal into a voice signal of the second speaker; a voice rule synthesis device that handles multiple languages;