JP2624958B2 - Speech synthesizer - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] In the present invention, speech is analyzed in advance in units of words, phrases, sentences, etc., and stored and registered in the form of feature parameters.
The present invention relates to a speech synthesizer for editing and synthesizing messages according to the contents of messages to be answered.

[Conventional technology]

FIG. 2 is a block diagram showing a basic configuration of a conventional speech synthesizer. In FIG. 2, reference numeral 1 denotes a speech signal a input at the time of registration, which is analyzed in a predetermined unit by a PARCOR analysis method or the like. 2 is a parameter storage circuit for storing the analysis characteristic parameter b output from the analysis circuit 1, and 3 is a parameter storage according to a message content c input from an input device (not shown). An editing circuit 4 for selecting a feature parameter from the circuit 2 and editing the obtained selected feature parameter d is a synthesis circuit 4 for generating a synthesized voice f from the editing feature parameter e from the editing circuit 3.

Next, the operation will be described. First, by inputting a plurality of types of voice signals a corresponding to a message to be answered in advance by uttering the same speaker, the analysis feature parameter b for each predetermined unit is extracted by the analysis circuit 1 and stored in the parameter storage circuit 2. Remember me. After the desired initial registration has been made in this way, every time a message content c to be answered is input, the editing circuit 3 retrieves the selected characteristic parameter d from the parameter storage circuit 2 in accordance with the message content c. The editing feature parameter e is synthesized by the synthesizing circuit 4, so that a synthesized voice f is output and a voice response is made.

[Problems to be solved by the invention]

However, in the conventional speech synthesizer of this type, when the content of a message to be responded to is changed, and the feature parameter of the stored and registered speech is partially corrected or added, the message is not registered by the initially registered speaker. The synthesized speech is partially unnatural due to the different speakers, and even if registered by the same speaker, it may also be partially affected by different microphones or other equipment or physical condition. A great difference in sound quality occurred, causing quality deterioration. As described above, the conventional voice response device has a problem that it is difficult to change a part of the registered voice.

The present invention has been made in order to solve the above problems, even when the speaker at the time of partial registration is different from the initially registered speaker, or when the same speaker uses different equipment and physical condition,
It is an object of the present invention to provide a speech synthesizer capable of obtaining a natural synthesized speech with constant quality during synthesis.

[Means for solving the problem]

A speech synthesizing apparatus according to the present invention includes a clustering circuit that generates a pseudophoneme pattern of an initial registration voice based on a feature parameter stored in a parameter storage circuit at the time of initial voice registration, and a pseudophoneme pattern that stores the generated pseudophoneme pattern. The storage circuit compares the feature parameters obtained from the analysis circuit at the time of partial registration with the onomatopoeia pattern stored in the onomatopoeia pattern storage circuit, and corrects the feature parameter with the onomatopoeia pattern having the highest similarity. And a matching circuit for generating the onomatopoeia expression feature parameters and storing the onomatopoeia expression feature parameters in a parameter storage circuit.

[Action]

According to the present invention, a pseudophoneme pattern of the initially registered speech is generated by the clustering circuit from the characteristic parameters obtained at the time of the initial registration, and stored in the onomatopoeia pattern storage circuit. At the time of partial registration such as addition or correction, the characteristic parameters obtained from the analysis circuit are corrected by the onomatopoemic pattern having the highest similarity in the onomatopoeia pattern storage circuit via the matching circuit and stored in the parameter storage circuit. Is done.

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of a speech synthesizer according to an embodiment. The same reference numerals as those in the conventional example of FIG. 2 denote the same or corresponding parts, and a description thereof will be omitted.

In the figure, reference numeral 5 denotes clustering in the parameter space (eg, the method of Gray et al., Literature; Linde Y., Buzo A., Gr
ay RM “An AIgorithm for Vector Quantizer Desig
n ", IEEE Trans Com., Vol. COM-28, No. 1, PP. 84-94 (Ja
n.1980), the clustering feature parameter g composed of a part or all of the analysis feature parameter b stored and registered in the parameter storage circuit 2 at the time of initial registration is calculated, and the feature of the initially registered speaker is calculated. A clustering circuit for generating an onomatopoeic pattern h that is representative of each phoneme, 6 a onomatopoeic pattern storage circuit for storing the onomatopoeic pattern h generated as described above, and 7 an analysis at the time of partial registration such as addition or correction. By comparing the analysis feature parameter b from the circuit 1 with the onomatopoeia pattern h stored in the onomatopoeia pattern storage circuit 6, the onomatopoeia pattern having the highest similarity is selected, and the above analysis feature parameter b is corrected. The matching circuit outputs the parameter to the parameter storage circuit 2 as the onomatopoeia expression feature parameter k. Although not shown, the analysis feature parameter b from the analysis circuit 1 is input directly to the parameter storage circuit 2 at the time of initial registration, and is input via the matching circuit 7 at the time of partial registration, by an operation of an operator or the like. Has become.

Next, the operation will be described. At the time of initial registration, a voice is input in units of words, phrases, etc., which are units of editing, converted into analysis characteristic parameters b by the analysis circuit 1, and then stored in the parameter storage circuit 2. Here, the characteristic parameter is generally a pitch period P and a vector representing the characteristics of the vocal tract. (For example, a PARCOR coefficient). In the following, a set of feature parameters It will be described later. At the end of the initial registration, a clustering feature parameter g composed of part or all of the analysis feature parameter b stored in the parameter storage circuit 2
Is sent to the clustering circuit 5, and a pseudophony pattern h is applied to the vector representing the characteristics of the vocal tract by, for example, the method of Gray et al. (I = 1, 2,... I), the average value of the pitch period is also calculated, and these are
Is stored.

In addition, at the time of addition or correction input, the analysis circuit 1 uses the analysis circuit 1 as an analysis feature parameter b. (Where j = 1,..., J) is extracted. Where qj is the pitch period, Is a vector quantity indicating the characteristics of the vocal tract, and j is an index indicating time. The analysis feature parameter b is sent to the matching circuit 7 and compared with the onomatopoeia pattern h stored in the onomatopoeia pattern storage circuit 6, and is used as the onomatopoeia expression feature parameter k. Is converted to

Where l (j) is (I = 1, ..., I) distance value Is the minimum value of i. Also, α
Is a coefficient for correcting pitch, for example, It is given. This onomatopoeia expression feature parameter k is added and stored in the parameter storage circuit 2 as correction data.

On the other hand, at the time of synthesis, the editing circuit 3 reads and edits the necessary selected characteristic parameter d from the parameter storage circuit 2 in accordance with the message content c as in the conventional case, and the synthesizing circuit 4 generates and outputs a synthesized voice f from the edited characteristic parameter e. I do.

〔The invention's effect〕

As described above, according to the speech synthesis apparatus of the present invention, a clustering circuit that generates a pseudophoneme pattern of an initial registration voice based on the feature parameters stored in the parameter storage circuit at the time of initial voice registration, and a generated pseudophoneme pattern And a feature phoneme pattern obtained by the analysis circuit at the time of partial registration and the onomatopoeia pattern stored in the onomatopoeia pattern storage circuit. A matching circuit that generates the onomatopoeia expression feature parameters by correcting the feature parameters and stores the onomatopoeia expression feature parameters in a parameter storage circuit, thereby enabling the difference in speaker or equipment or physical condition at the time of registration. Additions and corrections can be made while maintaining high quality by eliminating sudden changes in partial synthesized sound quality due to differences in sound quality, etc. In which exhibits an excellent effect of easily.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a speech synthesizer according to an embodiment of the present invention, and FIG. 2 is a block diagram showing an example of a conventional speech synthesizer. 1 is an analysis circuit, 2 is a parameter storage circuit, 3 is an editing circuit, 4 is a synthesis circuit, 5 is a clustering circuit, 6 is a pseudophoneme pattern storage circuit, and 7 is a matching circuit. In the drawings, the same reference numerals indicate the same or corresponding parts.

Claims (1)

(57) [Claims] An analysis circuit for analyzing a voice inputted at the time of registration in a predetermined unit to extract a feature parameter, and a parameter storage circuit for storing the extracted feature parameter,
A speech synthesizer for selecting and editing characteristic parameters from the storage circuit according to the content of a message to be answered, and synthesizing the edited characteristic parameters, wherein the initial registration is performed based on the characteristic parameters stored in the parameter storage circuit at the time of initial voice registration. A clustering circuit for generating an onomatopoeic pattern of the voice, an onomatopoeic pattern storage circuit for storing the generated onomatopoeic pattern, a feature parameter obtained from the analysis circuit at the time of partial registration, and stored in the onomatopoeic pattern storage circuit. A matching circuit that generates a pseudophoneme expression feature parameter by comparing the onomatopoeia pattern with the onomatopoeia pattern having the highest similarity and correcting the feature parameter, and stores the onomatopoeia expression feature parameter in a parameter storage circuit. A speech synthesizer comprising: