JP2584236B2 - Rule speech synthesizer - Google Patents

Description

The present invention relates to a speech synthesizer. In particular, the present invention relates to a speech synthesizer based on rules for synthesizing speech from arbitrary sentences.

(B) Conventional technology There has been active research on a rule-based speech synthesizer for synthesizing arbitrary sentences, and recently, trial production and commercialization of an optional sentence speech output device that can be connected to a newspaper review device or a personal computer have been performed. I have. When a sentence mixed with kanji or kana is input as a character code from a keyboard or the like to a rule speech synthesizer, a sequence of basic speech units and a symbol for performing prosody control are generated from those character strings. As the basic speech unit, a unit corresponding to a syllable such as a consonant-vowel chain (called a CV unit) or a unit of a consonant-vowel-consonant chain (called a CVC unit) is used. Symbols for controlling prosody include a symbol indicating an accent position and a symbol indicating a breathing position.In the case of Kana character input, in general, the symbol for performing prosody control is also used at the same time as the Kana character. , Input from outside.

On the other hand, audio data corresponding to each of the basic audio units is stored in a storage device (hereinafter, referred to as a basic audio data memory), and the corresponding audio data is read from the beginning according to the column of the basic audio units, The data is stored in the audio data buffer memory while being connected. Note that when audio data is connected, interpolation processing is performed at the connection portion.
In addition, a pause is inserted in the audio data or a fundamental frequency pattern is generated according to the symbol for performing the prosody control, and is added to the audio data in the audio data buffer memory.

The audio data in the audio data buffer memory generated by the above processing is converted into audio via an audio synthesis circuit such as LSP or PARCOR. Although the basic voice unit and the voice data in the basic voice data memory may correspond one-to-one, the characteristics of the voice data are considered to be different depending on the phonemic environment, for example, in the case of the beginning and the middle of the word. So, 1-6-19 of the Acoustical Society of Japan, October 1982
As shown in "Speech synthesis using CV syllables and V syllables", multiple types of speech data are prepared for the same basic speech unit, such as for the beginning, middle, and end, depending on the location in the sentence or word. In some cases, the sound quality is improved.

(C) Problems to be solved by the invention As in the former of the prior art, for the basic voice unit,
When only one type of voice data is prepared in the basic voice data memory, for example, the voice data in the basic voice data memory is created using naturally uttered voice uttered at the beginning of a word. In this case, among the basic speech units generated from the input sentence, those at the beginning of the sentence or word use the speech data generated in the same phoneme environment, so the sound quality is small, but the sentence At the end of a word, the sound quality is greatly degraded because the phonemic environment is different.

Also, as in the latter case of the prior art, in order to prepare a plurality of types of audio data for the same basic audio unit in order to prevent deterioration, a large capacity is required as a basic audio data memory. In the example of the above paper, 526 types of audio data were prepared, and 9600kbps PARCOR
When using parameters, it requires a memory capacity of 55KB.

(D) Means for Solving the Problems The present invention is to output an arbitrary sentence by voice,
Means for inputting a character string, basic sound data memory for storing sound data corresponding to each of basic sound units such as phonemes and syllables, means for generating a sequence of basic sound units from the input character string, and the basic sound unit Means for reading the audio data corresponding to the column from the basic audio data memory, an audio data buffer memory for sequentially storing the read audio data,
A voice synthesizing means for synthesizing voice from voice data stored in the voice data buffer memory; when reading voice data corresponding to a column of basic voice units, the basic voice unit exists other than at the beginning of a sentence or a word; In this case, there is provided an editing means for deleting a part of the head of the audio data and storing it in the audio buffer memory.

(E) Function Since the present invention is provided with such an editing means, it is possible to generate voice data suitable for a position in a sentence or word with a small amount of basic voice data memory. For example, using naturally uttered speech that is uttered at the beginning of a word,
If you create audio data in the basic audio data memory,
Deterioration is greater at the middle and end of words than at the beginning of a sentence or word. In comparison with the above, it is possible to suppress deterioration due to having only one type of audio data.

(F) Embodiment FIG. 1 shows an example of the configuration of a rule speech synthesizer that realizes the present invention.

When, for example, the word "Soyokaze" is input from the keyboard (1), the character string buffer pointer (3) is controlled by the pointer control unit (2), and the character string "Soyokaze" is stored in the character string buffer (4). Is stored in The character type determination unit (6) reads out the characters stored in the character string buffer (4) one by one, and if it is a kana character, closes the switch (5) and sets the code of the basic voice unit corresponding to the character to an address pointer. Set to (7). The address table (8) stores the location of the voice data corresponding to each basic voice unit in the basic voice data memory (10), and the address indicating the location is set in the address pointer (9). Is done.

On the other hand, if the character type determination unit (6) determines that the character is a kana character, the voice data read control unit (16) closes the switch (11) and corresponds to the basic voice unit in the basic voice data memory (10). Audio data is stored in the audio data buffer memory (12). At this time, if the beginning-of-word determining unit (17) determines that it is the beginning of a sentence or a word, it is read as it is. Otherwise, according to the contents of the leading deletion length table (15), it is read before reading. ,
The content of the address pointer (9) is increased by the audio data read control unit (16), and a part of the head of the audio data is skipped and stored in the audio data buffer memory.

As shown in FIG. 2, the head deletion length table (15) contains, for each basic voice unit, data to be skipped when the basic voice unit is present at the end of a sentence or word. The number (the number of frames) is stored. “FIG. 2 shows a part of the head deletion length table (15).” For example, for the input “Soyokaze”,
When the character "S" is input to the character type determination unit (6), since it is a Kana character, the code for the basic speech unit "S" is set in the address pointer (7). The address of the audio data corresponding to "SO" is stored in the address table (8), and the address is set in the address pointer (9). Here, since the beginning-of-word determining unit (17) knows that "so" is the beginning of the word, the voice data is directly converted to the voice data via the switch (11) according to the output of the voice data read control unit (16). Stored in the buffer memory (12). Next, for the character "Y", a code for the basic voice unit "Y" is similarly set in the address pointer (7), and "Y" is set.
Is set in the address pointer (9) in the basic voice data memory (10) of the voice data corresponding to the word "Y". In accordance with the output, the audio data reading control unit (16) reads the number of frames 6 (see the column "Y" in FIG. 2) from which the head of the audio data should be deleted from the head deletion length table (15), Address pointer by that value (9)
Add the value of. Therefore, the audio data corresponding to "Y" stored in the audio data buffer memory (12) via the switch (11) is deleted from the beginning by six frames.

A similar process is performed for "f", but since the value of the head deletion length table (15) is 0, the deletion is not performed, and the voice data corresponding to "f" is accumulated as it is, and With respect to the audio data, the audio data from which only six frames have been deleted from the beginning is stored in the audio data buffer memory (12).

FIG. 3 shows the audio data corresponding to each basic audio unit of “Soyokaze” and the audio data buffer memory (1
This is a diagram illustrating the audio data stored in 2). As shown in the figure, the voice data is connected while the data at the end of the word at the end of the word is deleted as indicated by the hatched portions, and is stored in the voice data buffer memory (12).

At the time of connection, interpolation processing is performed at the connection part of the voice data, and when the character type determination unit (6) finally determines that it is the end of the character string, the voice synthesis unit (13)
Is sent a voice utterance instruction signal. Upon receiving the voice generation instruction signal, the voice synthesis unit (13) synthesizes voice using the voice data in the voice data buffer memory (12), and outputs the voice as voice from the speaker (14).

In the above explanation, the head deletion length table (15),
Although the audio data read control unit (16) and the beginning-of-word determining unit (17) exemplify an editing unit for deleting the beginning of the basic audio data, the present invention is not limited to this. Can be deleted or added in place.

(G) Effect of the Invention As described above, in the rule speech synthesizer of the present invention, when speech data corresponding to a basic speech unit is connected to generate speech data for a sentence or word, Since part of the voice data is deleted in accordance with the location in the sentence or word, the deterioration of the synthesized voice can be suppressed.

Also, compared to the case where a plurality of types of audio data corresponding to each audio unit are prepared depending on the existence position, only one type is prepared, so that the memory capacity required for storing the voice data is greatly reduced. be able to. For example, only about 1/3 of the memory capacity is required as compared to the case where three types, ie, for the beginning, middle, and end, are prepared.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration example of a rule synthesizing apparatus for realizing the present invention, FIG. 2 is a memory diagram showing a part of a head deletion length table, and FIG. FIG. 3 is a schematic diagram illustrating a relationship between audio data corresponding to a basic audio unit and audio data stored in an audio data buffer memory. (1) Keyboard, (4) Character string buffer, (6) Character type determination unit, (10) Basic audio data memory, (12) Audio data buffer memory, (15)
... Head deletion length table, (16)... Voice data read control unit, (17).

Claims (1)

(57) [Claims] (1) To output an arbitrary sentence by voice,
Means for inputting a character string, basic sound data memory for storing sound data corresponding to each of basic sound units such as phonemes and syllables, means for generating a sequence of basic sound units from the input character string, and the basic sound unit Means for reading the audio data corresponding to the column from the basic audio data memory, an audio data buffer memory for sequentially storing the read audio data,
Voice synthesizing means for synthesizing voice from voice data stored in the voice data buffer memory, and when reading voice data corresponding to the column of the basic voice unit, the basic voice unit exists other than at the beginning of a sentence or a word. When performing the above, the rule speech synthesizer is provided with an editing means for deleting a part of the head of the speech data and storing it in a speech buffer memory.