JPS58134697A - Waveform editting type voice synthesizer - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a waveform editing type speech synthesis device, and more particularly to a speech synthesis device of the type that edits unit speech waveforms in units of phoneme chains such as consonant-vowel and vowel-consonant combinations. ◎ Voice is an extremely important method for conveying information from various mechanical devices to humans, and many voice synthesis devices or voice response devices have been proposed.

If the number of sentences or phrases to be synthesized is limited, this can be easily achieved by storing the natural sounds of the sentences or phrases and outputting them as needed.

Furthermore, if the desired sentence is composed of an unspecified number of words Φ, it is impossible to store the natural speech waveforms of all the words.

In such cases, it has conventionally been possible to memorize natural speech waveforms that are smaller phonological units than words, such as single syllables or phonological chains (consonant-vowel or vowel-consonant combinations, etc.). Then, by combining and editing these unit speech waveforms, the speech of the target sentence was synthesized.

This method has the problem of requiring a large-capacity storage device in order to add pitch accents to words and natural tones to sentences.

A conventional example that requires relatively little storage capacity is one that uses analysis and synthesis. This analyzes natural speech waveforms such as monosyllables that are the unit of editing, extracts and stores parameters representing its spectral shape, pitch, voicedness, etc., and uses the parameters representing the spectral shape during synthesis. controlling the transfer function of the synthesis filter based on the voiced case, inputting an impulse train with a period represented by a parameter representing pitch to the synthesis filter,
In the case of unvoiced speech, a desired speech waveform is generated by inputting irregular noise to the synthesis filter.

According to this method, the storage capacity required is smaller than when storing natural speech waveforms, and the pitch related to intonation and accent can be changed only by changing the parameters representing the pitch. It can be easily changed by □. However, the speech synthesized by this method has poor clarity and is not suitable for applications that require high-quality synthesized speech.

An object of the present invention is to provide a speech synthesizer of the type that edits and synthesizes unit speech waveforms in units of phonological chains such as consonant-vowel and vowel-consonant, etc., which produces relatively few synthesized speeches with relatively natural accents and tones. An object of the present invention is to provide a speech synthesis device that can be realized with limited storage capacity.

According to the present invention, there is provided a speech synthesis device of the type that edits unit speech waveforms in units of phoneme chains such as combinations of consonant-vowel and vowel-consonant, etc.; The present invention is characterized in that it includes converting means for converting the pitch of a waveform, and editing means for editing the pitch-converted waveform. A waveform editing type speech synthesis device can be obtained. Next, the present invention will be explained in detail with reference to the drawings. FIG. 1 shows a block diagram of one embodiment of the present invention.

・1.1 First, when a signal requesting the start of synthesis is input from the synthesis request input terminal 107 to the control circuit 105, the control circuit 105
Instructs the character string analysis circuit 102 via the first control signal transmission path 109 to input a character string and start analyzing the character string.

In accordance with the control signal sent from the control circuit 105, the character string analysis circuit 102 analyzes characters consisting of a character string representing a sentence to be synthesized and a control character string representing an accent or tone inputted from a character string input terminal 1.06. The column is converted into synthetic data information consisting of a data number of a unit speech waveform that is easy to synthesize and a pitch value to be converted if necessary by referring to an internal table, and the synthesized data information transmission line 113
The signal is sent to the control circuit 105 via the control circuit 105.

Next, the control circuit 105 receives the synthesized data information sent from the character string analysis circuit 102, and transmits the unit speech waveform data to the unit speech waveform storage circuit 101 via the second control signal transmission path 110. Specify the number and the destination of the unit audio waveform. In other words, when using the unit audio waveform as it is, the first
The waveform editing circuit 10 via the unit audio waveform transmission line 114 of
If it is necessary to convert one pitch, the signal is sent to the pitch conversion circuit 103 via the second unit audio waveform transmission line 115. Further, when it is necessary to convert the pitch, the control circuit 105 connects the third control signal transmission line 1
11, the pitch value to be converted is sent to the pitch conversion circuit 103 through the unit audio waveform storage circuit 10.
The pitch of the unit audio waveform sent from 1 is converted.

The pitch conversion circuit 103 converts the unit audio waveform storage circuit 101 according to the control signal sent from the control circuit 105.
It converts the pitch of the unit audio waveform sent from the unit and sends it to the waveform editing circuit 104 via the pitch conversion waveform transmission line 11'6.

Pitch conversion can be realized, for example, by expanding or contracting the waveform on the time axis. An example of a specific method will be described later.

Next, the control circuit 105 sends a control signal to the waveform editing circuit 104 via the fourth control signal transmission path 112, and when using a unit audio waveform, changes the pitch of the waveform sent from the unit audio storage circuit 101. When using the converted waveform, the waveform sent from the pitch conversion circuit 103 is input and edited, and 6- When input and editing of all waveforms are completed, synthesized speech is output from the synthesized speech output terminal 108.

Here, the principle of a method for expanding and contracting a waveform by digital signal processing will be explained with reference to FIG.

In FIG. 2(a), a perpendicular line m drawn as a solid line from the time axis indicates a sample value obtained when a waveform with a pitch period of 'l'po is sampled at a sampling period Ts of approximately -period. In order to convert the pitch, first, sample values are obtained by interpolating between the above sample values at every interpolation period TX. In FIG. 2(a), the interpolated sample values are indicated by vertical %1n drawn by a broken line from the time axis. Next, when these interpolated sample values n are arranged for each sampling period Ts, the pitch period of the waveform is converted to 'l'pl as shown in FIG. 2(b).

At this time, the interpolation period TI is α of the sampling period Ts
When Tr=α−Ts (1) holds, the converted pitch period TPI is 1/α of the pitch period Tpo of the original waveform, that is, the following equation (2). .

TP1=TPO/α (2) Therefore, the pitch frequency is multiplied by α. If α is smaller than 1, the pitch frequency will decrease, and if α is larger than 1, the pitch frequency will increase.

Note that the above explanation is only one example of a pitch conversion method, and other methods are also possible.

According to the present invention, although the storage capacity for a unit speech waveform is relatively small, by changing the pitch, a synthesized sound with a relatively natural accent and tone can be obtained.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of an embodiment of the present invention, and FIG. 2 is a diagram illustrating an example of a method for converting the pitch of an audio waveform. ,. 11□1 In Fig. 1, 101...Unit voice waveform memory 1 circuit, 102...Character string analysis circuit, 103...
... Pitch conversion circuit, 104 ... Waveform editing circuit, 105 ... Control circuit, 106 ... Character string input terminal, 107 ... Synthesis request input terminal,
108...Synthesized audio output terminal. 9- Figure 1

Claims (1)

[Claims] A speech synthesis device of a type that edits unit speech waveforms in units of phoneme chains such as combinations of consonant-vowel and vowel-consonant, etc., comprising a storage means for storing the unit speech waveforms, and a pitch of the unit speech waveforms. 1. A waveform editing type speech synthesis device comprising: a converting means for converting the pitch; and an editing means for editing the pitch-converted waveform.