JPH07199991A - Data generation device for speech synthesis - Google Patents

Abstract

PURPOSE:To automatically generate synthesized speech data for guidance broadcasting on a bus. CONSTITUTION:A tape on which broadcast contents are recorded is reproduced and its playback signal is converted into a digital signal, which is separated into basic speech patterns Bp by words and phrases by utilizing pause periods of the speech. It is checked whether or not the patterns Bp are registered in a data base 16, and when not, they are registered in the data base 16 together with given identification codes Ppi. A sequence of codes Bp corresponding to the basic speech pattern sequence of the input speech is generated and stored in a reproduction table 21 by input speeches.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a device for creating voice synthesis data in a voice synthesis broadcasting device used for in-vehicle broadcasting of a passenger car or the like.

[0002]

2. Description of the Related Art Japanese Patent No. 1504694 proposes a device for appropriately reading out a basic voice pattern such as a word or a phrase registered in advance in a car guidance broadcast of a shared vehicle and combining it to reproduce a desired broadcast voice. However, a method and a device for generating a basic voice pattern used in this device have not been proposed. Conventionally, a basic voice pattern is manually extracted and extracted from a series of broadcast voices, and it is discriminated each time whether the basic voice pattern is a variation pattern or a common pattern. For example, if the broadcast audio is "Next is Tokyo Station", "Next" and "Station" are common patterns.
“Tokyo” is extracted and discriminated as a fluctuation pattern. The basic voice pattern thus obtained is stored so as not to be duplicated, and the stored voice is edited to synthesize the broadcast voice.

[0003]

The conventional manual creation of voice-synthesized data is not so problematic when the amount of guide broadcasting is small, but when looking at the in-vehicle broadcasting of a shared vehicle, the number of routes and the number of stops are large. In this case, the amount of broadcast audio becomes long, and it becomes unrealistic to manually separate and extract the basic audio patterns and store them without duplication.

[0004]

According to the present invention, continuous speech is input by the input means, and the basic speech pattern at the word or phrase level is separated and extracted from the input speech by the pattern extraction means. The extracted basic voice pattern is determined by the determination means to be the first appearance, and when the determination is the first appearance, the registration means adds the identification code to the basic voice pattern and registers both. Also, for each input voice by the conversion means, using the extracted basic voice pattern and the registered identification code,
An identification code string of the basic voice pattern is generated.

[0005]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A case where the embodiment of the present invention is applied to the creation of voice synthesis data for in-car guidance of a shared vehicle will be described. The continuous sound to be broadcast is input as an electric signal by the input means. For example, broadcast audio is recorded on a magnetic tape in advance, the magnetic tape 11 is reproduced by a tape recorder 12, the reproduced output is converted into a digital signal by an AD converter 13, and the memory 1 such as a magnetic disk is used.
Stored temporarily in 4. For example, the manuscript of the guide broadcasting for the route of the shared vehicle shown in FIG. 2 is shown in FIG.
Is like. In this example, the route number is 9851, and the announcer reads this manuscript and records the voice on a magnetic tape. In this manuscript, characters such as "route number", "route name", "business broadcast: inside the car", "outside the car", and other names such as the stop name at the beginning of the sentence (indicating that it will be broadcast at the stop) are not read. Is also included.

[0006] When reading a manuscript, a pause having a different length depending on each word or phrase, between sentences, between a series of broadcast blocks and further between routes. Insert a period (voiceless period). For example, for each punctuation mark in the sentence, 1 is set as the resting period,
Insert 2 at the punctuation mark. The space in the manuscript is 1
A pause period 5 is inserted between a series of broadcast blocks for one stop guide and another block, and a pause period 8 is inserted between the blocks to separate the route.

When all the in-vehicle broadcasting voices are input to the memory 14 in this way, the basic voice pattern extracting means 1
In step 5, the stored contents of the memory 14 are sequentially read, and the basic speech patterns B _P at the word and phrase levels are sequentially separated and extracted with the pause period (non-voice section) as a clue. This extracted basic voice pattern B _P is the first one (unregistered one)
Whether or not is determined by the determination means 17 with reference to the basic voice pattern database 16. That is, the same basic voice pattern as the extracted basic voice pattern B _P is stored in the database 16
The basic voice patterns are compared by a pattern matching (for example, DP (Dynamic Programming) matching) method.

If the result of this determination is that it has not been registered, the identification code B _Pi is added to this basic voice pattern and it is registered in the basic voice pattern database 16 by the registration means 18. On the other hand, if it is already registered, the identification code B _Pg of the basic voice pattern is read from the database 16. For example, in the case of the example shown in FIG. 3, as shown in FIG. 4, the route number “9851” is registered as the identification number B _P 0, then “Showa garage front” is registered as the identification code B _P 1, and "Departure", "To Heisei housing complex", "This broadcast is", "Special 98
No. system ”... are identification codes B _P 2, B _P 3, B, respectively.
It is registered as _P 4, B _P 5, .... At this time, FIG.
In the case of the input voice of the manuscript of “For example, this broadcast is to Heisei housing complex No. 98 system” is registered as an identification code B _P 6 as a basic voice pattern as shown in FIG. Therefore, the basic voice pattern "I'm going to Heisei housing complex" extracted from the following input voice, "I'm sorry to keep you going to Heisei housing complex. We will refuse to bring dangerous materials." Is not given a new identification code.

Next, the conversion means 19 converts the input voice for each one broadcast sentence into an identification code string of the basic voice pattern in the arrangement order of the basic voice patterns, and stores it in the reproduction table 21 for each broadcast sentence. It This memory analyzes the length of the pause period used when separating and extracting the basic voice pattern to determine whether it is a word level break, sentence break, broadcast block break, or line break, and Create an identification code string for each.

[0010] Figure 3, with respect to speech "9851" as shown in FIG. 5 in the example shown in FIG. 4, the identification code B _P as a broadcast 1
Only 0 is given, and the voice "Showa garage front bound for Heisei housing complex"
On the other hand, as broadcast 2, identification codes B _P 1, B _P 2, B _P 3
A code string of is created. Similarly, a broadcast sentence is created. In order to perform voice synthesis using the voice synthesis data thus obtained, the reproduction data table 21 is sequentially read, and the basic voice pattern corresponding to each read identification code is read from the database 16 to perform voice synthesis. Good.

The basic voice pattern can be stored by compressing it by a differential coding method, an adaptive differential coding method or the like to reduce the storage amount. Further, the present invention is not limited to the guide broadcasting of a shared vehicle, but can be applied to the creation of voice synthesis data necessary for uttering other synthetic voices.

[0012]

As described above, according to the present invention, the input voice is automatically separated and extracted into the basic voice patterns, and the basic voice patterns formed in each of the series of the input voices are arranged in the order of the basic voice patterns. A sequence of identification codes indicating a voice pattern is generated, which is more efficient, faster, and more accurate than the case of manually creating such synthesized voice data, and even a long voice can be easily created. Moreover, there is no duplication that gives different identification codes to the same basic voice pattern.

[Brief description of drawings]

FIG. 1 is a block diagram functionally showing an embodiment of the present invention.

FIG. 2 is a diagram showing an example of routes of a shared vehicle.

FIG. 3 is a diagram showing an example of a guide broadcasting manuscript of a shared vehicle.

FIG. 4 is a diagram showing a storage example of a database 16 showing a relationship between a basic voice pattern and its identification code.

FIG. 5 is a diagram showing a storage example of a reproduction table 21.

Claims (4)

[Claims] 1. A means for inputting a continuous voice, a means for separating and extracting a basic voice pattern at a word or phrase level from the input voice, and determining whether the extracted basic voice pattern is the first appearance. And a means for registering by adding an identification code when the basic voice pattern is the first appearance in the determination, the input using the registered identification code and the extracted basic voice pattern. A voice synthesizing data creation device comprising: means for generating an identification code string of a basic voice pattern corresponding to a voice. 2. The input voice is provided with a pause period for each word, phrase, between sentences, and between a series of input voices, and the basic voice pattern is separated by using the pause period as a clue. The data synthesizing apparatus for voice synthesis according to claim 1, wherein 3. The length of the pause period for each word or phrase, the pause period between the sentences, and the pause period between the series of input voices are different from each other, and from the length of the pause period, 3. The voice synthesizing data creating apparatus according to claim 2, further comprising means for adding a code for distinguishing each series of input voices and storing the identification code sequence in the reproduction table for each series of input voices. . 4. The means for determining whether the first appearance is
The means for determining whether or not the separated and extracted basic voice pattern matches the basic voice pattern registered in the database by pattern matching. A voice synthesizing data creation device described in.