JPS62287299A - Voice responder - 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] 3. Detailed Description of the Invention (Field of Industrial Application) This invention analyzes speech in units of one word, nine sentences, etc. and stores and registers the results in advance in the form of feature parameters. The present invention relates to a voice response device that reproduces voice and responds by editing and synthesizing messages according to the contents of messages to be responded to.

[Prior Art] Fig. 2 is a block diagram showing the basic configuration of a conventional voice response device. CO1? An analysis circuit that performs analysis using the analysis method, etc., and extracts the feature parameter b representing the characteristics of the voice; 2 is a parameter storage circuit that stores the analysis '+ il & parameter b output from the analysis circuit 1; , 3 is an editing circuit that selects a feature parameter from the parameter storage circuit 2 in accordance with the message content C input from an input device (not shown), and edits the obtained selected feature parameter d; 4 is an editing circuit for editing the selected feature parameter d; This is a synthesis circuit that generates a synthesized speech f by performing an operation opposite to that of the analysis circuit 1 on the feature parameter e.

Next, the operation will be explained. First, by having the same speaker utter and input a plurality of types of audio signals a corresponding to the messages to be responded to, the analysis circuit 1 extracts analysis characteristic parameters b for each predetermined vehicle position, and stores them in the parameter storage circuit 2. Register it in memory. After the desired initial registration is made in this way, every time the message content C to be responded to is input, the field collection circuit 3 sends the message content C.
The selected feature parameter d is taken out from the parameter storage circuit 2 in accordance with the above and collected, and this edited feature parameter e
are synthesized by the synthesis circuit 4, and a synthesized voice f is output to provide a voice response.

[Problem that the invention seeks to solve]

However, in conventional voice response devices of this type, when the contents of a message to be responded to change and the registered voice characteristic parameters are to be partially modified or added, the initially registered speaker If you don't,
The synthesized voice will be extremely unnatural, with parts of the voice coming from different speakers, and even if the voice is registered by the same person, the quality of the voice will vary depending on the microphone or other equipment used, or depending on the person's physical condition. Differences occurred, causing quality deterioration. As described above, the conventional voice response device has a problem in that it is difficult to partially change the registered voice.

This invention was made to solve the above-mentioned problems, and even if the speaker at the time of partial registration is different from the initial registered speaker, or even if the same speaker uses different equipment or physical condition, the same level of performance is maintained during synthesis. The object of the present invention is to provide a voice response device that can obtain high-quality and natural synthesized speech.

[Means for solving problems]

The voice response device according to the present invention includes a clustering circuit that generates an onomatopoeic bait pattern of an initially registered voice based on the feature parameters stored in a parameter storage circuit at the time of initial voice registration, and an onomatopoeic decoy that stores the generated one pseudophonetic pattern. The pattern memory circuit compares the feature parameters obtained from the analysis circuit during partial registration with the onomatopoeic n-member pattern stored in the condensed phonetic pattern memory circuit, and corrects the feature parameters using the one pseudophonetic pattern with the highest degree of similarity. and a matching circuit for storing parameters in a parameter storage circuit.

[Effect]

In this invention, the clustering circuit generates a consonant rhyme pattern of the initially registered speech from the feature parameters obtained at the time of initial registration, and stores it in the condensate decoy pattern register. During partial registration such as additions or corrections, the feature parameters obtained from the analysis circuit are corrected by the onomatopoeic pattern with the highest degree of similarity in the onomatopoeic pattern memory circuit through the matching circuit, and are recorded in the parameter memory circuit. 4g is given.

[Example] Hereinafter, an example of the present invention will be described with reference to the drawings. 1st
The figure is a block diagram showing the configuration of the voice response device of the embodiment, and the same reference numerals as in the conventional example in FIG. 2 indicate the same or corresponding parts, and the explanation thereof will be omitted.

In the figure, 5 is a class kring in the parameter space (e.g. Gray et al.'s method 2 literature; Linde Y,
, Buzo A, , Gray R,M, “A.
n Algorithm for Vector Qua
ntizer Design”, IEEE Tr.
ansCom,, Vol, COM 28. No
, 1. PP, 84-94 (Jan.

1980)), clustering feature parameters g consisting of a part or all of the analysis feature parameters b stored and registered in the parameter storage circuit 2 at the time of initial registration.
6 is an onomatopoeia that stores the onomatopoeic pattern h generated in the above manner. A decoy pattern storage circuit 7 compares the analysis feature parameter b from the analysis circuit 1 with the onomatopoeic pattern h stored in the original sound ■ pattern storage circuit 6 at the time of partial registration such as addition or modification, and selects the one with the highest degree of similarity. This is a matching circuit that selects an onomatopoeic pattern, corrects the analysis feature parameter b using this pattern, and outputs it to the parameter storage circuit 2 as an onomatopoeic decoy expression feature parameter. Although not shown in the drawings, the information is input through the analysis circuit 7 during initial registration by an operator's operation or the like.

Next, the operation will be explained. At the time of initial registration, speech is input in units of two words, etc., which are the units of editing, and after being converted into analysis feature parameters b in the analysis circuit 1, the parameters are stored in the circuit 2.

Here, the characteristic parameters are generally the pitch circumference MP and the vector C1 representing the characteristics of the vocal tract <for example, PARCOI? coefficients, etc.). Below, a set of feature parameters will be expressed as (Pan). When the initial registration is completed, the clustering feature parameter g consisting of a part or all of the analysis feature parameter b stored in the parameter storage circuit 2 is sent to the clustering circuit 5, and the onomatopoeic silver pattern is created by, for example, the method of Gray et al. As h, Oli (
i=l, 2°·-N) are calculated, and the average value of the pinch period is also calculated, and these are stored in the onomatopoeic silver pattern storage circuit 6.

Furthermore, when inputting additions or corrections, the analysis circuit 1 inputs (qj; Ibj) as the analysis feature parameter b (where j =
1. -, J) are extracted. Note that qj is a pinch period, Toj is a vector L indicating characteristics of the vocal tract, and j is an index indicating time. This analysis feature parameter b is sent to the matching circuit 7, and the onomatopoeic silver pattern storage circuit 6
is compared with the onomatopoeic silver pattern h stored in
l).

Here, f(j) is ibj (!: OJi (i
= 1. -, I) which takes the minimum value for the distance value d (Toj, Gi). Also, α
is a coefficient for correcting pinch, and is given by, for example. This onomatopoeic silver expression characteristic parameter k is stored in the parameter storage circuit 2 as additional and modified data.

On the other hand, at the time of synthesis, the editing circuit 3 reads out and edits the necessary selection feature parameter d from the parameter storage circuit 2 according to the message content C, as in the past, and the synthesis circuit 4 generates and outputs the synthesized speech f from this edited feature parameter e. do.

[Effects of the Invention] As described above, the voice response device of the present invention includes a clustering circuit that generates an onomatopoeic silver pattern of an initially registered voice based on the feature parameters stored in a parameter storage circuit at the time of initial voice registration; The onomatopoeic silver pattern memory circuit that stores the onomatopoeic silver pattern, and the characteristic parameters obtained from the analysis circuit during partial registration are compared with the pseudophonetic pattern stored in the onomatopoeic silver pattern memory circuit, and the onomatopoeic silver pattern with the highest degree of similarity is compared. Equipped with a matching circuit that corrects the characteristic parameters using the onomatopoeic silver pattern and stores them in the parameter recording circuit, it is possible to prevent sudden changes in partial synthesized sound quality due to differences in speakers, equipment, physical condition, etc. at the time of registration. This has the excellent effect of allowing additions and corrections to be easily made while maintaining high quality.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing a voice response device according to an embodiment of the present invention, and FIG. 2 is a block diagram showing an example of a conventional voice response device. 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 an onomatopoeia pattern storage circuit, and 7 is a matching circuit. Note that the same reference numerals in the figures indicate the same or equivalent parts. Agent Masuo Oiwa (2 idiots) Procedural amendment (name of spontaneous 20 inventions Voice response device 3, person making the amendment Relationship to the case Patent applicant address 2-2-3 Chiyodazu Marunouchi, Tokyo) Name (601) Mitsubishi Electric Corporation Representative Shiki
Moriya 4, agent address 2-2-3-5 Marunouchi, Chiyoda-ku, Tokyo
, a column for detailed description of the invention subject to amendment. 6. Contents of amendment (1) In the 7th and 8th lines of page 2 of the specification, the phrase "response by playing audio" is amended to read "reply by audio." (2) In the 1st and 2nd lines of page 3 of the same book, the phrase ``synthesized speech f by applying the above-mentioned to e'' is corrected to ``synthesized speech f from e.'' . that's all

Claims (1)

[Claims] It is equipped with an analysis circuit that analyzes the voice input at the time of registration in predetermined units and extracts feature parameters, and a parameter storage circuit that stores the extracted feature parameters, and extracts the feature parameters from the storage circuit according to the content of the message to be responded to. In a voice response device that performs a voice response by selecting and editing and synthesizing the edited feature parameters, an onomatopoeic pattern of the initially registered voice is generated based on the characteristic parameters stored in the parameter storage circuit at the time of initial voice registration. The clustering circuit, the onomatopoeic pattern memory circuit that stores the generated onomatopoeic patterns, and the feature parameters obtained from the analysis circuit during partial registration are compared with the onomatopoeic patterns stored in the onomatopoeic pattern memory circuit. A voice response device comprising: a matching circuit that corrects the characteristic parameter using an onomatopoeic pattern having a high degree of similarity and stores the corrected feature parameter in a parameter storage circuit.