JPS63169700A - Unit voice editing type rule 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 (Field of Industrial Application) The present invention relates to a regular speech synthesizer, and more particularly to a regular speech synthesizer that generates a synthesized speech waveform from an input phoneme symbol string.

(Prior Art) Conventionally, in a regular speech synthesizer, there is a method of editing and synthesizing speech waveforms using cv, vc, etc. as unit speech. For example, Voice Study Group Material 882-06 (April 1982)
A technique described in "Arbitrary word synthesis method using pitch-synchronous interpolation of CV and VC waveforms" is known.

(Problem to be solved by the invention) The conventional method did not take into account the influence of the preceding and succeeding unit voices (articulatory combination), so although the clarity of individual voices was good, the synthesized voice sounded unnatural. Ta. In order to solve this drawback, it is natural to think of a method in which a plurality of unit speech waveforms corresponding to various articulatory coupling conditions are extracted and prepared in advance from natural speech for each cv and vc. However, in this method, the pitch and spectrum envelope may become discontinuous depending on how the unit speech waveforms are combined.

(Means for Solving the Problems) The regular speech synthesis device of the present invention includes means for decomposing an input phoneme string representing speech to be synthesized into unit phonetic name strings such as CV and VC, and means for calculating pitch frequency data according to a pitch rule from the unit phonetic name sequence; , a memory for storing boundary data of each unit voice such as formant, and a memory for storing each unit voice waveform (memory, and a memory for storing boundary data of each unit voice such as formant, etc.). For example, means for selecting unit speech names such that the sum of the distances between the preceding and succeeding unit speeches in one word is minimum, and means for editing the unit speech waveform and synthesizing speech according to the selected unit speech names. has.

(Operation) This method calculates allophone subsets from the input phoneme string, generated unit phonetic name strings such as CV and VC, and pitch frequency data generated by pitch rules, and selects the preceding and following allophone subsets from among the allophone subsets. An optimal unit speech sequence is selected in which the sum of the distances (functions of pitch, formant, etc.) between unit speeches is the minimum within a certain speech interval (for example, one word). In order to select the optimal unit speech sequence, the sum of the distances for all possible combinations of unit speech included in the allophone subset is determined, and the minimum one is selected. In this way, by incorporating the influence of the preceding and following phonemes and optimizing the continuity of pitch and formant between unit voices for synthesis, synthesized speech with high clarity and smooth sound quality can be obtained.

(Embodiment) FIG. 1 is a block diagram showing an embodiment for realizing the principle of the present invention. The phoneme string input from the input terminal 1 is input to the unit phonetic name string decomposition circuit 2 and the pitch rule circuit 3, where it is decomposed into unit phonetic name strings such as cv, vc, etc.
3, the pitch frequency value (with some width) of each unit voice is determined according to the phoneme sequence. Based on the results of steps 2 and 3, an allophone subset selection circuit 4 selects an allophone subset that is a set of a plurality of unit voices. The optimum value selection circuit 5 uses the pitch and formant data at the boundaries of each unit voice stored in the boundary condition storage memory 6 to determine the sum of the distances between the preceding and succeeding unit voices within a speech interval (for example, one word). Select from among the allophone subsets selected in step 4 so as to be the minimum. Using the optimal abnormal sound subset selected in step 5 and the unit speech waveform stored in the unit speech waveform memory 7, the speech synthesis circuit 8 edits and synthesizes the unit speech waveform and outputs it from the output terminal 9. Regarding the unit speech waveform editing and synthesis method, please refer to the above-mentioned Bunyu Speech Study Group fee 88.
2-06 (April 1982) "Arbitrary Word Synthesis Method Using CV, VC Filtered Pitch Synchronous Interpolation" is detailed, so the explanation will be omitted here.

(Effects of the Invention) As explained above, the present application uses pitches, formants, etc. that are optimized within a certain range (for example, one word) from unit phonetic name sequences such as cv and vc, pitch rules, and allophone subsets. By doing this, it is possible to obtain a synthesized sound that takes into account the effects of the preceding and following phonemes, resulting in high clarity and smooth sound quality.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the present invention. 1... Input terminal 2... Unit phonetic name string decomposition circuit 3... Pitch rule circuit 4... Allophone subset selection circuit 5...・
Abnormal sound subset optimal value selection circuit 6...Boundary condition memory memory 7...Unit speech waveform memory 8...Speech synthesis circuit

Claims (1)

[Claims] means for decomposing an input phoneme string representing speech to be synthesized into unit phonetic name strings such as CV and VC; means for calculating pitch frequency data from the input phoneme string according to a pitch rule; means for selecting an allophone subset according to pitch frequency data; a memory for storing boundary data of each unit voice such as pitch and formant; a memory for storing each unit voice waveform; Means for selecting a unit phonetic name from a subset using the boundary data so that the sum of the distances of preceding and following unit phonetics within a certain phonetic interval (for example, one word) is minimum; and the selected unit phonetic name. and means for editing the unit speech waveform and synthesizing speech according to the following.