JPH0477794A - Voice synthesizer - Google Patents

Abstract

PURPOSE:To present a voice synthesizer, which can effectively generate a synthesized voice with high naturality, by controlling whether the voice is made nasal or not and whether the voice is made long vowel or not, according to the information of a nasal sound or a long vowel sound when converting the information of reading shown by an accent dictionary to the sequence of vocal sounds. CONSTITUTION:According to the nasal sound information and the long vowel sound information to be applied from an input part 1, the application of nasal sound regulation and long vowel sound regulation at a vocal sound sequence certification part 4 is controlled, and there is difference in the sequence of vocal sounds to be generated there. Namely, it is controlled whether the vocal sound to be the object of the nasal sound is converted to the nasal vocal sound or not and whether the sound of (i) is replaced with the long vowel sound [-]or not, and the sequence of vocal sounds is generated corresponding to an input instruction. As the result, only by controlling the input of the nasal sound information and the long vowel sound information corresponding to the purpose of voice synthesization, the voice can be synthesized with high naturality corresponding to the purpose or locality.

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Industrial Application Field) The present invention relates to a speech synthesis device that can generate synthesized speech according to a purpose from a character code string with good naturalness.

(Prior Art) Recently, an input character code string is analyzed to obtain its phoneme sequence and prosodic information, and a phonological parameter string and a prosodic parameter string are generated by applying predetermined rules according to the phoneme sequence and prosody information. Various speech synthesis devices have been developed that generate synthesized speech based on these parameter sequences. A speech synthesis device based on this type of rule synthesis method has the advantage that it can relatively easily generate synthesized speech representing arbitrary words or sentences, compared to speech synthesis devices using conventional recording and editing methods. Therefore, in combination with voice recognition technology, it is attracting attention as an important technology for realizing highly natural man-machine interfaces.

By the way, this type of speech synthesis device is used, for example, to output sentences created using a word processor.
The scope of its use is expanding.

Recently, devices have also been devised to incorporate functions that allow, for example, changing the pitch of the voice and the rate of speech, so that the synthesized and output voice can be processed to the extent desired.

By the way, in the case of Japanese audio, the ga line [ga, gi, gu.

Ge 1 Go, Gear, Gyu, Gyo] is a nasalized sound [Power 9. Ki0. ri0 ke0ko0. Ki0ya,
Ki0yu, Kio=3] exists. In order to deal with such nasalized sounds, conventional speech synthesis devices create in advance speech segments for the G line that are not nasalized and speech segments for the G line that are nasalized. The information is registered in the accent dictionary. These speech segments are then selectively used to generate synthetic speech. for example[
Kagamiko is voice synthesized as C Shika 0 Miko.

Even outside the volume range, for example, [Bank] is synthesized as [Ginkoko].

However, this kind of nasalization does not always occur depending on the type of word that contains the G sound, and there are some words that are pronounced without nasalization depending on the linguistic region.

Also, regarding the A sound that follows the koretsu sound, for example, [teacher]
As typified by, there are cases where this is pronounced as [Sensei] and cases where it is pronounced as [Sensei]. In this case, for boat fishing, register the pronunciation [-] in the accent dictionary where it corresponds to [Iko].
We deal with the hikion by obtaining information about the ``sense'' reading.

However, if you register the accent in the accent dictionary,
There is a problem in that it is no longer possible to synthesize speech without converting this into a subtone. In other words, there is a problem in that it is not possible to arbitrarily obtain synthesized speech that is made into a pitched sound or synthesized speech that is not made into a pitched sound according to the user's preference.

(Problem to be Solved by the Invention) In this way, in the past, whether or not to make the speech to be synthesized nasalized or accentuated was determined based on rules registered in advance in an accent dictionary. There was a problem in that it was difficult for the user to arbitrarily bend the sound to obtain the desired sound with good naturalness.

The present invention has been developed in consideration of the above circumstances, and its purpose is to easily give instructions for nasalization and accentuation, and to effectively generate highly natural synthesized speech. The object of the present invention is to provide a speech synthesis device that can perform the following.

[Aside from the Invention (Means for Solving the Problems) The present invention analyzes an input character code string to obtain its phoneme sequence and prosody information, and uses the phoneme sequence and prosody information to calculate phonemes according to predetermined rules. The present invention relates to a speech synthesis device that generates a parameter string and a prosodic parameter string and generates synthesized speech based on these parameter strings, and in particular, a means for inputting nasalization information indicating whether or not to nasalize the synthesized speech to be generated. If the input nasalization information indicates nasalization of the voice, a predetermined nasalization rule is applied to obtain a phoneme sequence including a nasalized phoneme,
Further, when the nasalization information indicates non-pancreaticization, the phoneme sequence is determined without applying the cupration rule.

In addition, a means is provided for manually inputting hikionization information indicating whether or not to hikitonize the A sound that follows the gaku sound of the synthesized speech to be generated, and the input hikihonization information is converted into the hikionization of the A sound. , the A sound is generated instead of the A sound, and conversely, when the A sound is not changed, the A sound is generated as is. It is something to do.

(Function) According to the present invention, by inputting nasalization information specifying whether or not to enable the nasalization rule, it is possible to easily control whether or not to enable the nasalization rule, This makes it possible to selectively synthesize nasalized speech or non-nasalized speech. Furthermore, it is possible to easily control whether or not to convert the A sound that follows the second row of sounds into a raised sound and perform speech synthesis using the raised sound information. As a result, depending on the purpose of speech synthesis, you can simply control the nasalization information and the consonantization information according to the purpose of speech synthesis.
It becomes possible to synthesize speech with good naturalness to suit the purpose and regional characteristics.

(Embodiment) Hereinafter, a speech synthesis device according to an embodiment of the present invention will be described with reference to the drawings.

FIG. 1 is a schematic configuration diagram of an embodiment apparatus, and numeral 1 is an input section for inputting character code strings such as words and sentences. Via this human power section 1, nasalization information, accentuation information, etc., which are characteristic of this embodiment apparatus, are also input. The input character code entered manually from the human power section 1 is given to the word collation section 2 and is subjected to collation with the accent dictionary 3. Further, the nasalization information and the consonant information inputted manually from the input section 1 are provided to the phoneme sequence testing section 4. The word matching unit 2 matches the input character code string with an accent dictionary 3 in which accents, parts of speech, pronunciations, etc. of a plurality of words are registered in advance, and extracts accent information and part of speech information regarding the words for which a match is detected. is given to the accent type testing section 5, and information on the pronunciation of the word is given to the phoneme sequence testing section 4.

Therefore, the phoneme sequence testing section 4 converts the reading information about the input character code string given from the word matching section 2 into a phoneme series, but at this time, the nasalization information and the accentuation information given from the human power section 1 are used. The phoneme sequence is determined by performing different processing according to the categorization information. That is, when the nasalization information indicates the nasalization mode, the nasalization rule is applied to nasalize the sound to be nasalized, and the G sound is removed from the G sound at the beginning of the input character code string. into a nasalized phoneme. Furthermore, when the nasalization information indicates a non-nasalization mode, all the G sounds in the input character code string are converted into phonemes as they are. In other words, in the non-nasalization mode, the reading information is converted into phonemes without applying the nasalization rule.

In addition, in the phoneme sequence verification section 4, when the kinonization information given from the input section 1 indicates kinonization, the phonology sequence verification section 4 applies the kinonization rule to determine the sound that follows the kusunon in the reading information. When the A sound is changed to a Hirikon [-], and the above-mentioned Hirikonization information indicates non-Hirikonization, the A sound following the Gakuon is converted into a phoneme as it is.

On the other hand, the accent type testing section 5 determines the accent type of each accent phrase in accordance with one or more pieces of accent information provided from the word matching section 2 and its part of speech information.

However, the accent type determined in this manner is provided to the synthesis parameter generation section 7 along with the phoneme sequence phoneme-ized as described above by the phoneme sequence verification section 4. Then, the synthesis parameter generation section 7 generates a phoneme parameter string corresponding to the phoneme sequence by referring to the speech segment file 6, and also generates a prosodic parameter string according to the accent type information. The speech synthesis unit 8 generates synthesized speech according to the phonetic parameter string and the prosody parameter string generated in this manner, and outputs the synthesized speech.

That is, in this embodiment, the application of the nasalization rule and the consonantization rule in the phoneme sequence verification section 4 is controlled according to the nasalization information and the consonantization information provided from the human power section 1, and the phoneme sequence generated therein is controlled. There is a difference between the two. In other words, it is possible to control whether or not to convert the phoneme that is the target of nasalization into a nasalized phoneme, and whether or not to replace the i sound with a consonant [-], and to generate a phoneme sequence according to the input instruction. It has become.

According to this device configured in this way, for example, when the word "Corporation Corporation" is input from the input section 1, the input section 1 supplies the character code string "Corporation Corporation" to the word matching section 2. Then, the word matching unit 2 matches the input character code string with the accent dictionary 3 configured as shown in FIG. It is given to the phoneme sequence testing section 4.

When the nasalization information given from the input section 1 indicates the nasalization mode, the phoneme sequence test section 4 nasalizes the "power" of the above reading "kabushiki kaisha" and pronounces it as r kabus.
Convert to vik1f'aisjaJ. However, [s v
i] indicates a devoiced "shi", and r raJ indicates a nasalized "ga". The phoneme sequence converted in this way is given to the synthesis parameter generation section 7, which generates a phoneme parameter sequence corresponding to the phoneme sequence with reference to the speech unit file 6.

On the other hand, the accent type testing section 5 verifies that the accent type of "Corporation Corporation" is [Type 5] from the accent dictionary 3. According to such accent type information,
The synthesis parameter generation section 7 generates the prosodic parameter sequence. Then, the speech synthesis unit 8 generates and outputs the synthesized speech as "Kabushikika 0 Isha" according to the phonetic parameter string and the prosody parameter string generated as described above.

On the other hand, even if the same input character code string "Co., Ltd." is given, if the nasalization information given from the human resources department 1 indicates a non-nasalization mode, the phoneme series test Section 4 converts the pronunciation "Kabushiki Riki Isha" directly into r kabusvikigaisja J. As a result, by following the phoneme parameter sequence created based on such a phoneme sequence, the speech synthesis unit 8
Now we will generate and output a synthesized voice called "Kabushiki Gaisha".

To explain another example, if the word ``kirei'' is input as an input character code string and the hikion mode is specified, for example, if the word ``kirei'' is determined from the accent dictionary 3 shown in FIG. '' is directly converted into a phoneme sequence rklrelJ by the phoneme sequence verification section 4.

However, [1] indicates a draw sound. Since the accent type for the input word of lever is determined as [type 1] from the accent dictionary 3, the speech synthesis section 8 generates and outputs the synthesized speech "Kirei".

On the other hand, if the non-biki-on mode is specified, the "-" in the above pronunciation "kire-" is changed to the hiki-on [-] as the A sound that follows the second-row sound "shi". Therefore, by returning this to the original sound "i", the phonetic series rklrelJ
generate. As a result, the speech synthesis unit 8 determines the synthesized speech as "beautiful" according to the phoneme blur meter string based on the above-mentioned phoneme sequence.

According to the present apparatus configured and operated in this way, it is possible to easily select and control whether or not to nasalize the speech to be synthesized by inputting the nasalization information. In addition, by inputting the pull sound information, the A sound that follows the second row of sounds is either synthesized as "i" or the sound is synthesized as the pull sound "-".
It is possible to easily select and control whether to convert it into speech and synthesize it.

Therefore, without changing the configuration (content) of the accent dictionary 3, it is possible to easily select and control nasalization and accentuation to synthesize and output a desired voice. Moreover, by simply instructing whether or not to apply the nasalization rule based on the nasalization information, and whether or not to apply the nasalization rule based on the nasalization information, it is possible to make the synthesized speech nasalized and accentuated very easily. This has great practical effects, such as making it possible to control soundization.

Note that the present invention is not limited to the embodiments described above. For example, the contents and structure of the accent dictionary 3 are not limited to the example described above. Also, register the nasalized reading information in the accent dictionary, and change it to the original reading (
It is also possible to return it to the sound part) and use it for speech synthesis. Of course, it is also possible to register in the accent dictionary 3 information on pronunciations that are not converted into accented sounds. In addition, the present invention can be implemented with various modifications without departing from the gist thereof.

[Effects of the Invention] As explained above, according to the present invention, when converting reading information shown in an accent dictionary into a phoneme sequence, it is possible to determine whether or not to perform nasalization according to nasalization information or elongation information. Furthermore, since it is possible to control whether or not to make the sound more pronounced, great practical effects can be achieved, such as the ability to synthesize and output a desired voice very simply and effectively.

[Brief explanation of the drawing]

The figures show a speech synthesis device according to an embodiment of the present invention. FIG. 1 is a schematic configuration diagram of the embodiment device, and FIG. 2 is a diagram showing an example of the configuration of an accent dictionary in the embodiment device. DESCRIPTION OF SYMBOLS 1... Input part, 2... Word collation part, 3... Accent dictionary, 4... Phonological series test part, 5... Accent type test part, 6... Speech element file, 7 ...Synthesis parameter generation unit, 8...Speech synthesis unit. Applicant's agent Patent attorney Takehiko Suzue (11 wo Kajuki◆No. 3 [!Sound Information Bunsei Kou Ts1) (Aisei Kansen Figure 1)

Claims (2)

[Claims] (1) Analyze the input character code string to obtain its phoneme sequence and prosodic information, generate a phoneme parameter string and a prosodic parameter string from this phoneme sequence and prosodic information according to predetermined rules, and based on these parameter strings. A speech synthesizer that generates synthesized speech by using a voice synthesizer, comprising means for inputting nasalization information indicating whether or not to nasalize the synthesized speech to be generated, and when the nasalization information indicates that the voice is nasalized. A sound characterized in that a phoneme sequence is determined by applying a predetermined nasalization rule, and when the nasalization information indicates non-nasalization, a phoneme sequence is determined without applying the nasalization rule. Synthesizer. (2) Analyze the input character code string to obtain its phonological sequence and prosodic information, generate a phonological parameter sequence and a prosodic parameter sequence from this phonological sequence and prosody information according to predetermined rules, and generate a phonological parameter sequence and a prosodic parameter sequence based on these parameter sequences. A speech synthesis device that generates synthesized speech by using a voice synthesizer, comprising a means for inputting tone conversion information indicating whether or not to convert the A sound following the E sound of the synthesized speech to be generated into a tone, If indicates that the A sound has been made into a drawn sound, a drawn sound is generated instead of the A sound, and when the said drawn sound information indicates that the A sound is not made into a drawn sound, the A sound is generated as is. A speech synthesis device characterized by: