JPH06337700A - Voice synthesizer - Google Patents

Abstract

PURPOSE:To easily generate a synthesized sound exceedingly similar to a voice of a speaker. CONSTITUTION:This device is constituted of a voice input means 10 inputting the voice of the speaker, a recognition processing means 11 receiving an input voice and recognizing the input voice based on a prescribed acoustic segment network, an acoustic segment network storage means 13 storing the acoustic segment network corresponding to every word becoming an object of the synthesized sound, an acoustic segment network updating means 12 comparing and deciding the characteristic of the input voice and the acoustic segment network, and updating and registering the acoustic segment network according to the comparison result, a character input means 14 inputting the voice of the speaker as the word by a character line, a synthesis processing means 15 synthesizing the input word by a prescribed voice synthesis processing based on the acoustic segment network and a synthesis sound output means 16 outputting a synthesized sound.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesizer which outputs synthetic speech similar to that of a user (speaker). The present invention can finally be applied to an information processing device having a voice dialogue function.

[0002]

2. Description of the Related Art As a conventional voice recognition device, Japanese Patent Laid-Open No.
No. 3-223793 “Registered voice input / output device”,
There are Japanese Unexamined Patent Publication No. 63-183498 “Registered voice input / output device” and Japanese Unexamined Patent Publication No. 61231600 “Voice recognition device”. Is for outputting the voice of the registered speaker as a synthesized voice, converting the voice of the speaker into a feature parameter in advance, calculating the similarity between the input voice and the registered voice at the time of use, and registering the feature parameter registered in advance. The one most similar to is output as a synthetic sound of the recognition result. The similarity is determined by the well-known PARCOR analysis.

In the above, PARCOR analysis is used, but in this example, ADPCM synthesis is used to determine the degree of similarity. Is capable of performing voice registration on behalf of a specific speaker, and automatically performs registration processing of a registered word of the specific speaker in advance with a voice synthesized by a voice synthesizer.

FIG. 4 is a block diagram of an example of a conventional voice synthesizer using a sound segment network. In the figure, 41 is an acoustic segment network storage unit, 42 is a character input unit, 43 is a synthesis processing unit, and 44 is a synthesized sound output unit. In this method, a plurality of acoustic segment networks are stored in advance in the acoustic segment network storage unit, and when a voice is input from a speaker, the acoustic segment network that most resembles the voice is extracted and a corresponding synthesized voice is generated. Is.

[0005]

However, the above-mentioned conventional method is a method of extracting a network most similar to the speaker's voice from a plurality of acoustic segment networks that have been facilitated in advance. There are still many problems in the similarity of. An object of the present invention is to easily generate a synthetic voice that is very similar to the voice of the speaker.

[0006]

FIG. 1 is a block diagram showing the principle of the present invention. As shown in the figure, the present invention is a voice synthesizing apparatus for generating a synthetic voice similar to a voice of a speaker, which receives voice input means 10 for inputting the voice of the speaker and input voice from the voice input means. A recognition processing means 11 for recognizing the input voice based on a predetermined acoustic segment network, an acoustic segment network storage means 13 for storing an acoustic segment network corresponding to each word to be a target of a synthetic voice, and characteristics of the input voice. Acoustic segment network updating means 12 for extracting and comparing the characteristic with the acoustic segment network stored in the acoustic segment network storage unit, and updating and registering the acoustic segment network based on the characteristic of the input voice based on the comparison result.
A character input means 14 for inputting the voice of the speaker as a word in a character string, and a synthesis processing means 15 for synthesizing the input word from the character input means by a predetermined voice synthesis processing based on the acoustic segment network, And a synthesized voice output means 16 for outputting synthesized speech, extracting a feature of a speaker's voice, storing the feature in the acoustic segment network storage means, and then inputting a character string input from the character input means. When the synthesizing means synthesizes the sound by a predetermined speech synthesizing process, a synthesized sound is generated according to the updated acoustic segment network stored in the acoustic segment network storing means, and as a result, the speech of the speaker It is characterized in that similar synthetic sounds are generated.

Further, the acoustic segment network updating means 12 includes a feature extraction unit 22 for extracting acoustic features of a voice input by a speaker, an extracted segment and an acoustic segment network stored in the acoustic segment network. And a feature registration unit 24 that updates and registers the acoustic segment network based on the determination result of the feature determination unit.

Further, an input analysis unit 312 for analyzing a sentence input by a speaker by voice, a knowledge base 35 for storing knowledge about a task, and a predetermined reference to an input from a speaker with reference to the knowledge base. It further includes a sentence generator 38 that generates a sentence.

[0009]

According to the present invention, when a speaker inputs a voice, for example, by a voice input means such as a microphone, an acoustic segment network of the voice is stored, while a character input means such as a keyboard is used to input the voice of the speaker. When a character string of a voice is input as a word, a synthesized voice similar to the voice of the speaker can be easily generated according to the stored acoustic segment network.

[0010]

FIG. 2 is a block diagram of an embodiment of the present invention. 20
Is a voice input unit, 21 is a recognition processing unit, 22 is a feature extraction unit,
Reference numeral 23 is a feature determination unit, 24 is a feature registration unit, 25 is an acoustic segment network storage unit, 26 is a character input unit, 27 is a synthesis processing unit, and 28 is a voice output unit.

The voice input unit 20 is, for example, a microphone, the character input unit 26 is, for example, a keyboard, and the voice output unit 28 is, for example, a speaker. Speech recognition processing unit 2
1 recognizes a voice pattern from a microphone. The feature extraction unit 22 extracts the acoustic feature of the user's voice. The feature determination unit 23 compares the network of words stored in the acoustic segment network storage unit 25 with the acoustic features extracted by the feature extraction unit 22, and determines the difference. The feature registration unit 24 stores the features obtained by the feature determination unit 23 in the acoustic segment network. The acoustic segment network storage unit 25 stores the acoustic segment network of the word group to be recognized or synthesized. The synthesis processing unit 26 converts the character string input from the keyboard into synthetic sound.

The operation of this embodiment having such a configuration will be described below. The acoustic segment may be a phoneme unit or a syllable unit. These are connected according to rules so as to show various changes in pronunciation of each word to be subjected to voice recognition or synthesis, and after the acoustic segment network is generated, it is registered in the acoustic segment network storage unit 25.

Next, when a certain user recognizes a word input by voice, the feature extraction unit 22 extracts the acoustic feature of the voice of the user, and the result is extracted by the feature determination unit 23.
Pass to. The feature determination unit 23 compares the result with the network of words stored in the acoustic segment network storage unit 25 and determines whether there is a difference. If there is a difference, the feature registration unit 24 updates the network by registering it in the acoustic segment network storage unit 25.

By converting the character string input by the user with the keyboard into the synthetic sound, the synthesis processing unit 27 outputs the synthetic sound by using the updated acoustic segment network stored in the acoustic segment network 25. , Generates a synthetic sound close to the user's voice. FIG. 3 is a block diagram of another embodiment of the present invention. This configuration shows a voice dialogue system. In the figure, 30 is a microphone as a voice input unit, 31 is a voice recognition processing unit, 32 is a CRT as a display unit of the recognition result, 33 is a feature extraction unit for extracting acoustic features of the user's voice, and 34 is a sound. A feature determination unit that compares the network of words stored in the segment network storage unit 36 with the acoustic features extracted by the feature extraction unit 33 and determines the difference between them is denoted by 37. The feature is determined by the acoustic segment network storage unit 36. The feature registration unit 36, the acoustic segment network storage unit 36 that stores the acoustic segment network of the word group to be recognized or synthesized, 39 the keyboard that is the input means of the word character string, and 310 that is input from the keyboard. A synthesis processing unit for converting a character string into a synthesized voice, 311 is a speaker that is a means for outputting a synthesized voice, and 35 is a topic. A knowledge base that stores knowledge about a task, 38 is a sentence generation unit that generates a predetermined sentence using the knowledge base in response to a user input, and 312 is an input sentence analysis unit that analyzes a sentence that the user has input by voice. Are shown respectively.

The task may be, for example, a summary of an international conference, a reservation for a Shinkansen ticket, or a domestic travel guide. The knowledge base expresses and stores the knowledge about such tasks, for example, the information about the travel packages for the Tohoku area departing from December in a rule, which analyzes the user's voice question and provides an appropriate answer. Gives the system the ability to generate as a sentence.

[0016]

As described above, according to the present invention,
It is possible to easily generate a synthesized voice that is very similar to the voice of the speaker.

[Brief description of drawings]

FIG. 1 is a principle configuration diagram of the present invention.

FIG. 2 is a configuration diagram of an embodiment of the present invention.

FIG. 3 is a configuration diagram of another embodiment of the present invention.

FIG. 4 is a diagram illustrating a conventional example configuration.

[Explanation of symbols]

 10, 20, 30 ... Voice input unit 11, 21, 31 ... Recognition processing unit 12 ... Acoustic segment network updating unit 13, 25, 36 ... Acoustic segment network storage unit 14, 26, 39, 42 ... Character input unit 15, 27 , 43, 310 ... Synthesis processing unit 16, 28, 44, 311 ... Synthetic sound output unit 22, 33 ... Feature extraction unit 23, 34 ... Feature determination unit 24, 37 ... Feature registration unit 35 ... Knowledge database 38 ... Text generation unit 312 ... Input sentence analysis unit

Claims (3)

[Claims] 1. A voice synthesizer for generating a synthetic voice similar to a voice of a speaker, comprising voice input means (10) for inputting the voice of the speaker, and input voice from the voice input means. Recognition processing means (11) for recognizing the input voice based on a predetermined acoustic segment network; acoustic segment network storage means (13) for storing an acoustic segment network corresponding to each word to be the target of the synthesized speech; An acoustic segment for extracting a feature of an input voice, comparing the feature with an acoustic segment network stored in the acoustic segment network storage unit, and updating and registering the acoustic segment network based on the feature of the input voice based on the comparison result. Network updating means (12) and character input means (14) for inputting a speaker's voice as a word in a character string. A synthesis processing means (15) for synthesizing an input word from the character input means by a predetermined speech synthesis processing based on the acoustic segment network, and a synthetic sound output means (16) for outputting the synthesized speech. The feature of the voice of the speaker is extracted, the feature is stored in the acoustic segment network storage means, and then the character string input from the character input means is synthesized by the synthesis processing means by a predetermined voice synthesis processing. At this time, a synthetic sound is generated according to the updated acoustic segment network stored in the acoustic segment network storage means, and as a result, a synthetic sound similar to the voice of the speaker is generated. Voice synthesizer. 2. The acoustic segment network updating means (12) stores a feature extraction unit (22) for extracting acoustic features of a voice input by a speaker, the extracted features and the acoustic segment network. A characteristic determination section (23) for comparing and determining existing acoustic segment networks,
The voice synthesizing apparatus according to claim 1, further comprising a feature registration unit (24) that updates and registers the acoustic segment network based on the determination result of the feature determination unit. 3. An input analysis unit (312) for analyzing a sentence input by a speaker by voice, and a knowledge base (35) for storing knowledge about a task which is information such as recent reservations and guidance.
The speech synthesis apparatus according to claim 1, further comprising: a sentence generation unit (38) that refers to the knowledge base in response to an input from a speaker and generates a predetermined sentence.