JP2586040B2 - Voice editing and synthesis device - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a voice editing / synthesizing apparatus used for a voice response system.

(Prior Art) Conventionally, there has been known a method of storing a voice waveform of a word or a sentence uttered by a human, and performing a voice response by editing and synthesizing the voice waveform. Also, CV and VC
(Where C represents a consonant and V represents a vowel), a voice response system that edits and synthesizes a relatively short voice segment according to a character string given as an input and synthesizes an arbitrary voice is published by The Acoustical Society of Japan in 1982. Symposium of the Study Group (Document No. -S82-0)
6 (1982-4)), which is known from a document and the like entitled "arbitrary word synthesis method by pitch synchronous interpolation of CV and VC waveforms".

(Problems to be Solved by the Invention) However, in the former method, since the natural sound to be edited has a relatively long time length, the synthesized speech edited and synthesized has good sound quality, but the types of sentences that can be synthesized are limited. It has the disadvantage of being. Further, the latter method has a drawback that, although an arbitrary sentence can be synthesized, the synthesized speech quality is relatively inferior because the time length of a speech unit to be edited is short and the effect of articulation coupling is not sufficiently considered. doing.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a speech editing / synthesizing apparatus capable of generating an arbitrary sentence speech of relatively high quality while considering the influence of articulation coupling as much as possible.

(Means for Solving the Problems) The present invention provides a voice data memory that stores voice data such as words in advance together with a phoneme name string and syllable boundary data representing each of the voice data, and a syllable name string given as input. Means for matching a syllable name string (including a partial string) corresponding to the word or the like and selecting a partial syllable name string of the voice data that matches the longest, and a syllable name string selected by the selecting means Means for extracting desired audio data from the audio data by using the syllable boundary data and editing and synthesizing the audio data to generate a desired audio.

(Principle of the Invention) The change characteristics of characteristic parameters such as the frequency spectrum of syllables in speech such as continuously uttered words and sentences are as follows.
It is known that there is a great difference due to the influence of syllables before and after comparison with the change characteristic of the characteristic parameter of a syllable uttered independently, and this is called articulation coupling. In a rule-based speech synthesis system in which a plurality of unit speeches are cut out from natural speech in advance and prepared, and these unit speeches are edited to synthesize an arbitrary speech,
In order to enhance the synthesized sound quality by sufficiently considering the effect of the articulation coupling, it is necessary to prepare the unit voice as long as possible (having a large number of syllables). However, if the unit speech is long, the number of syllable combinations becomes enormous, which makes it difficult to cut out the unit speech from natural speech and also increases the scale of the speech synthesis system. Therefore, as a method capable of outputting relatively high-quality synthesized speech for the scale of the speech synthesizer, a method of adding word speech data including a frequently used syllable sequence together with a segmentation indicating a syllable boundary may be used. Second
In the figure, as an example, a syllable name (here, CV and VC are called syllables .: C is a consonant, V is a vowel) and a boundary data (segmentation data) for the word / yamazaki /. In order to enhance the effectiveness of this method, it is necessary to represent a syllable sequence included in a sentence to be synthesized with a syllable sequence as long as possible included in the word. Can be.

Note that, as the audio data, an audio waveform or a formant parameter extracted from the audio waveform can be used. A method for synthesizing an arbitrary sound from a sound waveform corresponding to a syllable (CV, VC) is described in, for example, the above-mentioned literature, and a method for synthesizing an arbitrary sound from a formant parameter or the like corresponding to a syllable is described in, for example, 1985 Of the Symposium on Symposiums (Document No. S85-31 (1985-7)) titled "Formant, CV-VC Type Rule Synthesis", and will not be described here.

(Example) An example of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the present invention. In this embodiment, first, a character string 110 representing a sentence to be synthesized is converted into a syllable name sequence 101 via a character string input terminal 11 and input to the optimum syllable sequence selection unit 2.
The optimal syllable sequence selection unit 2 refers to the syllable name sequence corresponding to the word voice data stored in the syllable name sequence storage unit A in the storage unit 3 and converts the syllable name sequence into partial syllables of the word voice data. It is decomposed into name strings 102 and output as address data of the syllable boundary data storage unit B in the storage unit 3. In the decomposition operation into the partial syllable name sequence, the syllable name sequence converted from the input character string is represented by (S ₁ , S ₂ ,... S _i , S _{i + 1} ,..., S _n ,.
S _N ) (where S _i ,... S _n are syllable names, and N is the number of syllables for the input character string), starting from S ₁ and extending from left to right in the longest part included in the word voice data Search syllable name string. As a result, assuming that S ₁ ,..., S _i are the longest partial syllable name sequence, the same operation as above is repeated from S _{i + 1} to sequentially search for the longest partial syllable name sequence. From the voice data storage unit C in the storage unit 3, the voice data 103c of the voice part corresponding to the partial syllable name string included in the word voice data is obtained according to the partial syllable name string of the word voice data.
Are sequentially output to the edit / synthesis circuit 4. The editing / synthesizing circuit 4 edits and synthesizes the audio data 103c output from the storage unit 3 and outputs it via a synthesized waveform output terminal 12 which generates a synthesized waveform.

In the above description, the presence or absence of the accent in the syllable name string was not taken into consideration, but the accent also affects the frequency spectrum pattern of the syllable. It is clear that a syllable sequence with relatively good synthesized sound quality can be obtained by performing the search with the addition.

(Effect of the Invention) As described above, according to the present invention, a relatively long syllable sequence can be used as a unit voice. Therefore, by adopting the speech editing / synthesizing apparatus of the present invention, it is possible to generate an arbitrary synthesized speech having a relatively high quality by incorporating the effect of articulation coupling more.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention, and FIG. 2 is a diagram showing an example of a syllable name string and boundary data. In FIG. 1, 1 is a syllable name string conversion unit, 2 is an optimal syllable sequence selection unit, 3 is a storage unit, 4 is an editing and synthesis circuit, 11 is a character string input terminal, 12 is a synthesized waveform output terminal, and A in the storage unit 3 is a storage unit of a syllable name sequence corresponding to voice data such as a word, B is a storage unit of syllable boundary data corresponding to the syllable name sequence in the voice data, and C is a storage of the voice data. Represents each part. In FIG. 2, the horizontal axis represents time, the vertical axis represents the average amplitude value, () represents a syllable name, and the solid line represents the average amplitude pattern of the word voice / yamazaki /.

Claims (2)

(57) [Claims] 1. A voice data memory for storing voice data such as words in advance together with a phoneme name sequence representing each voice data and syllable boundary data, a syllable name sequence given as input and a syllable name sequence corresponding to the word etc. Means for selecting a partial syllable name string of the voice data that matches the longest, and using the syllable boundary data in accordance with the syllable name string selected by the selecting means, converts necessary voice data from the voice data. Means for generating a desired voice by cutting out and editing and synthesizing. 2. The speech editing / synthesizing apparatus according to claim 1, wherein said phoneme name string stored in said speech data memory includes a partial string.