JPH04162000A - Voice compounding method by borrowing consonant - Google Patents

Abstract

PURPOSE:To reduce necessary memory capacity by calling and combining voice data of a consonant part and a vowel part corresponding to the consonant part to compound a voice waveform. CONSTITUTION:Voice data of a consonant part, on every column as for syllables other than five vowels 'A'-'O' and 'N', and a vowel part of all syllables are individually registered in a syllable table of Japanese, and the voice data of the consonant part and the vowel part corresponding to the consonant part are called and combined to compound a voice waveform. Respective vowel parts in the syllable table are not commonly used but separately registered respectively, and a resembled consonant part is commonly used (borrowing consonant). This can compress a voice data quantity and also prevent the lowering of voice quality.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a method for synthesizing Japanese speech. To explain in more detail, when synthesizing syllables in the same row on the syllable table,
The present invention relates to a voice synthesis method using borrowed consonants, which compresses the amount of voice data by combining and synthesizing one of the consonant parts of the row with a selected vowel part of the row.

The present invention is a very useful technology for a speech synthesis output device that outputs various messages, and can improve the quality of speech and reduce memory capacity.

[Prior Art] One of the Japanese speech synthesis methods is the waveform encoding method. This registers audio in units of sentences, phrases (words), and syllables (50 syllables), and combines them to reproduce and output various words. The basic method for registering these audio signals in a form close to the original audio signal is PCM (pulse code modulation), and D is a more efficient method.
There are PCM (differential pulse code modulation) and ADPCM (adaptive differential pulse code modulation).

Does the method of registering speech waveforms in sentences or words have the advantage of being able to play back the intonation of the recorded voice as is?However, there is a limit to the number of registered sentences and words, so only regular words can be played back. The disadvantage is that it cannot be used and has a limited range of applications.

On the other hand, when it comes to syllables and units (50 syllable units), it is a hassle to assign intonation to each syllable, and it is difficult to express beautiful intonation as in the method described above, or there is no limit to the number of sentences that can be reproduced. Therefore, its applications are expected to become more widespread in the future.

If sounds are registered in units of syllables, the total number of registered sounds will be over a hundred or so, depending on the registration method, such as direct sounds (clear sounds and voiced sounds) and persistent sounds. The syllables are all made up of consonants and vowels, except for the five vowels ``AJ'', ``Saij'' and ``N''. Figures 1 and 2 show the speech waveforms of the ``su'' and ``shibo'' sounds in the ``su'' line.As shown, the consonant part appears first, and the vowel part or It consists of a part, a pitch part, and an envelope.

Next, Table 1 shows a syllable table (the 50-syllabary table is expanded to include voiced sounds and persistent sounds).

(Hereafter in the margin) Table 1 (hereinafter in the margin) There are two types of consonants in the syllables from ``yugori'' to ``t''. When you pronounce that line with the consonant of "Yu", as shown in Table 1, it becomes the sounds of "Yu", "T", "Tou J", "R Te J", "T J", and "Chi" and "Tsu". There are different consonants.

Similarly, if you pronounce the Da line with the same consonants, it becomes "Da" and "Dei".
, "do", "de", "do". In the syllable table, ``ji'' and ``zu'' are pronounced the same as ``ji'' and ``zu'' in the za line.

From the above, the number of consonants used in Japanese syllables can be summarized into 27 types. The vowels are "a", "i",
There are only five types: ``tsu'', ``ko'', and ``o''. Simply guessing, all Japanese words can be uttered by combining 33 sounds in total: 5 types of vowels, 27 types of consonants, and ``n''. This method is actually used in low-quality speech synthesis output devices.

[Problems to be Solved by the Invention] Although this speech synthesis method can reduce the memory capacity for registering speech waveforms, the quality is low and the generated speech is no more than a mechanical sound.

If a speech synthesis output device using a monosyllable registration method can maintain a certain degree of quality, all the syllables listed in Table 1 or something similar thereto will be registered and used. However, this increases the memory capacity; for example, even in the case of the smallest device with 8-bit PCM at a sampling frequency of 10 MHz, the required memory capacity reaches 128 bytes.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a speech synthesis method that eliminates the drawbacks of the prior art as described above, reduces the required memory capacity, and maintains the same quality of speech.

[Means for Solving the Problems] The present invention, which can achieve the above object, has a system in which, on the Japanese syllable table, for syllables other than the five vowels of "AJ" to "OJ" and "N", one for each line. In this method, audio data for consonant parts and vowel parts of all syllables are registered separately, and a speech waveform is synthesized by calling and combining the consonant parts and the corresponding vowel part audio data. In other words, in the present invention, in order to pronounce any syllable in one line, one consonant part representing that line is combined with an individual vowel part, and since the consonant parts are commonly used, It is called "speech synthesis".

The present invention is based on syllable registration format PCM, ADPCMSDPC
M is applied to devices that reproduce PCM sound at the final stage of reproduction.

[Effect] The reason why the voice quality is inferior in the voice synthesis method that registers 33 sounds of vowels and consonants is that each vowel in each row of the syllable table (an expanded version of the 50-syllable table) is affected by its own consonant. . For example, the vowel of ``power'' (-a) and the vowel of ``su'' (-a) are clearly different. On the other hand, consonants within the same line of the syllable table are often very similar.

The present invention focuses on this point. That is, in the present invention, instead of using each vowel part in the syllable table in common, each vowel part is registered separately, and similar consonant parts are commonly used (
By doing so, the amount of audio data is compressed and the quality of the audio is prevented from deteriorating.

[Examples] Examples of the present invention will be described below. On the Japanese syllable table (a syllable table that expands the 50-syllabary table to include voiced sounds and sulky sounds), the five vowels "a" to "sai" and the syllable "n" are registered as they are. For other syllables,
For each line, audio data for one consonant part and vowel parts of all sounds are individually registered. For example, regarding the syllables in the ``su'' line shown in FIG. 1, only one arbitrary consonant part of each syllable from ``su'' to ``so'' is registered as the consonant part of the ``su'' line. The vowel parts (-a) to (-o) of each syllable in the second row are all registered separately.

Then, the selected consonant part and the data of the corresponding vowel part are recalled and combined to synthesize a speech waveform. During playback, for example, if you want to say "suji" (sushi), the consonant of the line "su", the vowel (-u) of the row "su", the consonant of the row "su", the vowel of the row "su", the consonant of the row "su", the sound (-i) )=Synthesize and pronounce in combination with "shi".

In an actual device, more intonation is added to make the words speak with higher quality.

[Effect of the invention] “As mentioned above, the present invention uses the consonant part in common for each row of the syllable table for each syllable other than the five vowels and “n”, so the amount of audio data can be compressed accordingly. . In other words, the required memory capacity can be reduced.

Since the consonant parts in the same line are often very similar, the voice quality is unlikely to deteriorate even if they are used in common as described above.

When converting audio waveforms into data using PCM, the amount of data becomes a problem in relation to the capacity of the semiconductor memory that stores it. According to the present invention, the memory capacity can be made as small as possible without deteriorating the voice quality, which directly helps to reduce the cost of the voice synthesis output device.

[Brief explanation of drawings]

FIG. 1 is a speech waveform diagram of the "s" sound, and FIG. 2 is a speech waveform diagram of the "sh" sound. Patent applicant Iwaki Electronics Co., Ltd.

Claims (1)

[Claims] 1. On the Japanese syllable table, for the five vowels "a" to "o" and syllables other than "n", the audio data of one consonant part for each line and the vowel part of all syllables are individually recorded. A voice synthesis method using borrowed consonants, characterized in that voice data of a consonant part and a vowel part corresponding to the consonant part are registered and combined to synthesize a voice waveform.