JPH0667685A - Speech synthesizing device - Google Patents

Abstract

(57) [Abstract] [Object] The present invention relates to a voice synthesizer for converting a text consisting of a character string into voice and uttering the voice, and a voice in which the utterance ends exactly at a predetermined time without losing the naturalness. The purpose is to realize a synthesizer. In a voice synthesizing device for reading a character string of a document, which comprises a phoneme converting unit 1, a parameter calculating unit 2, a voice synthesizing unit 3, and a voice waveform synthesizing unit 4, an input means 10 for designating a utterance time, The expansion / contraction processing unit 11 and the expansion / contraction processing unit are provided.
Reference numeral 11 changes the parameter relating to time calculated by the parameter calculation unit 2 so that the reading of the character string is performed at the designated utterance time.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesizer for converting text consisting of character strings into speech and uttering the speech.

With the development of multimedia technology, it has become possible to output a document by voice, but a more natural utterance and an easier specification method are required in the field of presentation. Specifically, it is required to specify the speaking time instead of the speaking rate.

[0003]

2. Description of the Related Art FIG. 4 is a block diagram of a conventional system. The phoneme conversion unit 1 converts the input document file written in kanji or kana-moji text into a kana character string by using a phoneme dictionary to form a phoneme string. This corresponds to the reverse operation of kana-kanji conversion in a word processor.

To this, accent information is added to a phoneme sequence forming a word by an accent dictionary, and further pause time information and intonation information in consideration of connection strength due to connection between words are added by grammatical analysis. , A composite character string.

The synthetic character string includes katakana that represents syllables, delimiters that represent exhalation paragraphs, phrases, and accent phrases,
It consists of accent marks, nasalization marks, unvoiced marks, etc.
The parameter calculation unit 2 performs the preprocessing of the speech synthesis unit 3 as follows.
Converts a composite string into a numeric parameter. In this,
The duration (time parameter) of each phoneme element is included.
This is the value at the standard speaking rate.

The voice synthesizing unit 3 transforms and synthesizes numerical parameters according to a certain synthesizing rule (for example, PARCOR method) to obtain parameter time series data. The parameter time-series data is an array of values for each fixed time unit (frame), and the voice waveform synthesizer 4 converts this into a voice waveform.

FIG. 5 shows an example of data at each processing stage from conversion of input text to numerical parameters. Figure 5
(A) is the input text. This is processed by the phoneme conversion unit 1, and the resultant synthetic character string is shown in FIG. 5 (B). As a result of processing this by the parameter calculation unit 2, FIG.
Numerical parameters at the previous stage of the parameter time series data as shown in (C) are obtained. In the figure, the column of pause represents silent time, the column of c represents consonant time, and v represents vowel time.

[0008] Numerical parameters are processed by the voice synthesizing unit to form parameter time-series data, and the parameter time-series data is processed by the voice waveform synthesizing unit 4 to obtain voice as analog data and output as voice from a speaker or the like.

When it is desired to change the utterance time, the utterance speed is changed by changing the frame time in the voice waveform synthesizer 4.

[0010]

In the prior art, the utterance speed could be changed, but the utterance could not be finished at a predetermined time.

It is an object of the present invention to realize a voice synthesizing apparatus in which the time is specified for the entire document or for each sentence so that the utterance can be finished exactly at a predetermined time without losing the naturalness.

[0012]

FIG. 1 is a block diagram showing the principle of the present invention. The conventional speech synthesizer is provided with an input means 10 for designating a utterance time and an expansion / contraction processing section 11 for changing a time parameter among numerical parameters.

[0013]

The utterance time when uttering with the standard time parameter is obtained by summing the time parameters of each phoneme element. Assuming that this is t and the designated utterance time is T, T / t is the expansion / contraction ratio, and the expansion / contraction processing unit 11
Is multiplied by the time parameter of each phoneme element to obtain a stretched time parameter. By passing a numerical parameter including the expansion / contraction processing time parameter to the voice synthesizing unit 4, the utterance time can be adjusted to the designated time.

By the way, when a human utters, when the utterance speed is changed, the utterance does not change uniformly. For example, when the voice is divided into a voiced section and a silent section, the expansion / contraction rate is in the silent section. It has been empirically known that the expansion / contraction ratio varies depending on the type of the voice component, as is large.

Therefore, in order to maintain the naturalness, it is desirable to determine the value of the time parameter in consideration of the difference in the expansion / contraction ratio depending on the type of the component, instead of changing the entire expansion / contraction ratio at the same.

The standard utterance time is divided for each type of component, and the sum of the respective time parameters is t1, t2, t.
3, ..., t = t1 + t2 + t3 + ... The expansion and contraction rate for each type of component is 1 + k1, 1 + k2, 1
+ K3, ..., and their mutual relations are defined.

For example, k1: k2: k3: ...
If a: b: c: ..., then the designated utterance time is: T = (1 + k1) t1 + (1 + k2) t2 + (1 + k3) t3 + ... = t + k1.t1 + k2.t2 + k3.t3 + ··· = t + k1 · t1 + b · k1 · t2 / a + c · k1 · t3 / a ··· t + k1 (a · t1 + b · t2 + c · t3 + ···) / a Therefore, k1 = (T -T) ・ a / (a ・ t1 + b ・ t2 + c ・ t3 + ・ ・ ・ ・) k2 = b ・ k1 / a k3 = c ・ k1 / a ・ ・ ・ ・ ・ ・ ・ ・ ・The expansion / contraction rate for each type of element is obtained.

Therefore, depending on the type of each component, 1 +
The time parameters after expansion / contraction processing are obtained by multiplying k1, 1 + k2, 1 + k3, ... By the time parameters of each phoneme element. Further, as shown in FIG. 2, the overall expansion / contraction rate T / t is 1
The farther away from, the greater the variation of the expansion / contraction rate depending on the type of component. Therefore, the above values of a, b, c ...
Make a table for t, and use that value for k1,
By calculating k2, k3, ..., expansion / contraction with more naturalness can be performed.

[0019]

EXAMPLES Examples of the present invention will be described below. An example of calculation when the above-mentioned example “a telegram arrived at a bungalow early in the morning” was doubled and halved is shown.

The sum of the standard time parameters is 363
5, the types of components are "silence", "vowels", and "consonants"
When divided into three, t1 = 1100, t2 = 196
5, t3 = 570.

If a: b: c = 7: 3: 0, then T = 7000 (about twice) is specified. k1 = (7000-3635) / 7 / (7/1100 + 3/1965 + 0.57)
0) = 1.733 k2 = 0.743 k3 = 0 Therefore, the expansion / contraction rate is 2.73 times for silence, 1.74 times for vowel, and 1 time for consonant.

FIG. 3 (A) shows the time parameter changed by these expansion / contraction rates. When T = 2000 (about 1/2 times) is specified. k1 = (2000-3635) / 7 / (7/1100 + 3/1965 + 0.57)
0) =-0.842 k2 = -0.361 k3 = 0 Therefore, the expansion / contraction rate is 0.16 times silence, 0.64 times vowel, and 1 time consonant. Here, when the expansion / contraction rate becomes smaller than a certain value, it is set to that value. For example, the expansion / contraction rate of silence is set to 0 as the minimum value.

FIG. 3B shows the time parameter changed by these expansion / contraction rates. The utterance time may be specified as the time of the entire document or may be specified as a sentence unit. When specifying in text units, the time specification command may be embedded in the document.

It is more effective if the presentation time is designated as the time of the entire document, and the sentence to be particularly emphasized can be designated individually.

[0025]

As described above in detail, according to the present invention, it is possible to realize the document reading apparatus which can be set within the designated time. Also, by specifying the reading time, the naturalness of utterance is maintained even if the utterance speed is changed.

[Brief description of drawings]

FIG. 1 is a principle block diagram of the present invention.

FIG. 2 shows a relative change in expansion / contraction ratio for each type of audio component.

FIG. 3 shows an example of time parameters after expansion / contraction processing according to an embodiment of the present invention.

FIG. 4 is a configuration block diagram of a conventional system.

FIG. 5 is an example of data from input text to conversion into numerical parameters.

[Explanation of symbols]

 1 phoneme conversion unit 2 parameter calculation unit 3 speech synthesis unit 4 speech waveform synthesis unit 10 vocalization time designating unit 11 expansion / contraction processing unit

Claims (3)

[Claims] 1. A phoneme conversion unit (1) which converts a document into a kana character string or the like into a phoneme string by a phoneme dictionary, adds accent information, pause time information, intonation information, etc., and forms a synthesized character string, and A parameter calculation unit (2) that converts a character string into a numerical parameter that includes a time-related parameter at the time of standard utterance, and a combined character string that is converted into a numerical parameter is converted and combined according to a certain rule to create parameter time series data. In a voice synthesizer for reading a character string of a document, which comprises a voice synthesizer (3) for converting the parameter time series data into a voice waveform synthesizer (4) for converting into an analog voice signal, The input / output unit (10) for designating and the expansion / contraction processing unit (11) are provided. A speech synthesizer characterized by changing a parameter to read a character string at a designated utterance time. 2. For each type of constituent elements of the uttered voice,
The speech synthesizer according to claim 1, wherein a rate of change of a parameter relating to time is changed. 3. The speech synthesizer according to claim 2, wherein the ratio of the rate of change of the parameter relating to time is changed for each type of voice constituent elements in accordance with the ratio of the standard utterance time and the designated utterance time.