JP2001282268A - Speech data delivery system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable home page readers and telephone users to listen to information on weather forecasts, etc., as speech because a speech synthesis server 12 receives the text data of the information on the weather forecasts, etc., from an information source 14 and forms synthesized speech data by using, for example, CHATR in accordance with therewith and delivers the data to Web servers 16 and telephone servers 18 which are delivery destinations. SOLUTION: The respective servers are able to provide diversified services without the need for discretely installing speech synthesizers to the Web servers and telephone servers and without increasing running costs.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice data distribution system, and in particular, distributes, for example, weather forecasts, stock price information, road information, race results such as horse races, and professional baseball bulletins to a WWW server or a telephone server as voice data. Related to audio data distribution.

[0002]

2. Description of the Related Art Conventionally, it is possible to access a web server (homepage) via the Internet and acquire data or information as described above.

[0003]

However, in any of the conventional systems, such data or information is character information, and the user must read it with his / her eyes, which causes some inconvenience. Remains. In order to solve this problem, the data and information may be provided by voice from the homepage.

[0004] In order to transmit voice data from a homepage, it is necessary to provide a voice conversion function in a web server, which increases the financial burden on the homepage operator.

Therefore, a main object of the present invention is to provide a new audio data distribution system for distributing audio data to a web server or the like.

[0006]

A voice data distribution system according to the present invention comprises: a receiving unit for receiving text data from an information source; a voice synthesizing unit for generating synthesized voice data based on the text data; An audio data distribution system including a distribution unit for distributing to a distribution destination.

[0007] Preferably, the voice synthesizing means is CHATR.
(Spontaneous Utterance Speech Waveform Connection Method) For this purpose, the speech synthesis means includes a speech waveform database and a speech unit selection unit for reading a speech waveform signal from the speech waveform database according to index information corresponding to a phoneme string of text data.

[0008] The synthesized speech data thus synthesized is preferably delivered to a storage system of a computer existing on a network. The computer may be a web server or a telephone server.

[0009]

For example, text data such as weather forecasts, stock price information, road information, race results such as horse races, and professional baseball bulletins are received from an information source. This text data is
It is stored in the text data area of the memory. In accordance with the instruction of the scheduler, the voice synthesizing unit reads out the text data from the text data area, and creates the synthesized voice data according to the text data by the CHATR. This synthesized voice data is stored like voice data in a memory. The distribution means reads the audio data in the audio data area according to the instruction of the scheduler, and transmits the audio data to a distribution destination such as a web server indicated by the scheduler.

[0010] If the voice synthesizing means using the CHATR synthesizes voice data using an actual voice waveform of an arbitrary person, for example, a talent or announcer, the voice data is distributed in a human voice desired by the distribution destination. can do.

[0011]

According to the present invention, information such as a weather forecast can be output as voice data from a home page simply by receiving the voice data, for example, by a homepage operator or the like, which is very convenient for the user. Become. In addition, it is not necessary to separately install a speech synthesizer at a distribution destination such as a homepage operator, and various services can be provided with a small operation cost.

[0012] Also, the use of CHATR makes it possible to deliver extremely natural voice data synthesized with an arbitrary human voice.

The above objects, other objects, features and advantages of the present invention will become more apparent from the following detailed description of embodiments with reference to the drawings.

[0014]

Referring to FIG. 1, a voice data distribution system 10 according to this embodiment includes a voice synthesis server 12, which includes a weather forecast source 141, a stock price information source 142, and a road information source 143. And text data of a weather forecast, stock price information, road information, a race result, and a professional baseball bulletin from a race result source 144 and a professional baseball bulletin source 145 (hereinafter sometimes collectively referred to as an “information source 14”). These information sources 14 are typically information providers, but may be public institutions such as the Meteorological Agency or the Road Authority if possible.

In this embodiment, the speech synthesis server 12
This performs speech synthesis using CHATR,
It does not exclude other methods of speech synthesis. Then, the speech synthesis server 12 converts the text data received from the information source 14 into speech data, and
(Web) Supply or distribute to the server 16 or the telephone server 18. These are called "delivery destinations".

As an example of the distribution method, in this embodiment,
The speech synthesis server 12 directly writes to a predetermined storage area of a storage system (not shown) including a hard disk or the like of the servers 16 and 18. In this case, in the storage system of each of the servers 16 and 18, an area where the synthesized voice data is to be written is fixedly formed.

Each of the web servers 16 operates a home page. When a user accesses the home page via the computer network 20, the web server 16 can listen to synthesized speech data distributed from the speech synthesis server 12. .

Similarly, the telephone server 18 provides a cellular phone and other telephone services. When a user accesses the telephone server 18 via the telephone carrier 22, the synthesized voice from the voice synthesis server 12 is transmitted through the telephone. You can listen to the data.

The speech synthesizing server 12 sends each server 1
In distributing the synthesized voice data to 6 and 18, actually, a necessary data compression technique is appropriately adopted, and the compressed voice data is distributed. Also, in the embodiment of FIG. 1, the distribution is shown through a dedicated line.
Of course, this may be distributed via the computer net 20. In this case, it is needless to say that appropriate settings such as encryption and password are required.

The block diagram shown in FIG.
2, the speech synthesis server 12 includes a receiving computer 24, which receives text data from the information source 14 shown in FIG. Speech synthesis server 12 further includes a synthesis computer 26, which performs speech synthesis with CHATR, which will be described in some use later. The distribution computer 28 distributes the synthesized speech data created by the synthesis computer 26 to each of the servers 16 and 18 (FIG. 1). These computers 24, 26 and 28 are also computers, schedulers 30
The above-mentioned operation is performed according to the instruction of (1).

The text data received by the receiving computer 24 is stored in a text data area 32a of a memory 32 such as a hard disk.
The synthesized voice data created by the synthesis computer 26 is stored in the voice data area 32b of the memory 32. Therefore, the synthesis computer 26 reads the specified text data from the text data area 32a and writes the synthesized voice data to the voice data area 32b according to the instruction of the scheduler 30. Then, the distribution computer 28 reads the instructed audio data from the audio data area 32b according to the instruction of the scheduler 30, and transmits it to the instructed destination.

In the embodiment of FIG. 2, the synthesizing server 12 is shown as being constituted by several computers, but it is a matter of course that this may be constituted by one computer.

The details of the synthesizing computer 26 are described in, for example, JP-A-10-49193 [G10L5 /
04, 3/00], synthesized voice data can be created using the voice of a person actually spoken.

The synthesizing computer 26 shown in FIG. 2 can be functionally represented as shown in FIG. Referring to FIG. 3, synthesis computer 26 includes a voice waveform database 34 that stores actual voice waveform signal data of a person to be output as a synthesized voice. Is analyzed to generate a phoneme symbol sequence, phoneme alignment, and extract feature parameters. At this time, the audio waveform database 3
4 to the voice analysis unit 38, the text database 36
Speech waveform signal data in accordance with the phoneme notation (orthographic text) is given. In the phoneme alignment processing, the start point and the end point of each phoneme are determined according to the phoneme HMM (Hidden Markov Model) 40.

The feature vector or feature parameter extracted by the voice analysis unit 38 is stored in the feature parameter memory 4.
2 is stored. The characteristic parameters include a phoneme label, a start time (start position) of the phoneme in each file in the audio waveform signal database 34 in the memory 42,
Information such as fundamental frequency, phoneme time length, power, stress, accent, position with respect to prosodic boundary, and spectrum inclination is stored for each file specified by the index number.

The weight coefficient learning unit 44 determines the optimum weight coefficient while learning it. In order to select, from the speech waveform database 34, the most suitable sample for the acoustic and prosodic kankyo of a given target speech, it is first necessary to determine which features and how much contribute to the phonetic and prosodic names. You have to decide according to the differences. This is because the types of important feature parameters change depending on the properties of phonemes. For example, the speech fundamental frequency is extremely effective in selecting voiced sounds, but has little meaning in selecting unvoiced sounds. In addition, the effect of the fricative sound for a sound machine varies depending on the types of phonemes before and after. Therefore, the weighting factor learning unit 44 determines, for example, by using a linear regression analysis, how much weight is given to each feature in order to select an optimal phoneme.
Is automatically determined by. Then, the weight coefficient vector obtained by the weight coefficient learning unit 44 is stored in the memory 46 and used by the voice unit selection unit 48.

The voice unit selector 48 receives text data (input phoneme sequence) stored in the memory 32 from the receiving computer 24, selects a voice unit based on the phoneme sequence, and indexes the voice unit in the voice waveform database 34. Print the number. That is, the speech unit selection unit 48 should concatenate the phoneme string represented by the text data to be converted into speech data with the target cost representing the approximate cost between the target phoneme and the candidate phoneme. A search is made for a phoneme candidate sequence having the minimum cost including a connection cost representing an approximate cost between adjacent phoneme candidates, and the index number, the start position (time) and time length of each phoneme are provided to the speech synthesis unit 50. .

The voice synthesizer 50 accesses the voice waveform signal database 34 based on the index number, the start position (time) and time length of each phoneme, and outputs digital voice waveform signal data of the phoneme candidate. Then, it is stored in the audio data area 32b shown in FIG.

As described above, the text data is converted into synthesized speech data by the synthesis computer 26 using the CHATR. In this CHATR, as described above, the voice waveform signal data of the actual speaker stored in the voice waveform signal database 34 is used. Therefore, a very natural voice can be synthesized as compared with voice synthesis using a general pulse and white noise.

On the other hand, in the CHATR, if the audio waveform signal data is not strictly managed, the audio waveform signal data of a celebrity or a celebrity may be used without permission. In other words, in the CHATR, if a person's real voice is acquired for a certain period of time, a voice waveform signal database can be created. 12 (FIG. 1) expands the possibility of using the data other than the data to be provided, and the risk of unauthorized use of the voice of another person dramatically increases.

Therefore, in the present invention, the synthesized speech data is provided from the synthesis server 12 to a necessary distribution destination. For this reason, even if voice synthesis using CHATR is used in the present invention, voice waveform signal data can be strictly and unitarily managed, and such unauthorized use can be completely avoided. In addition, in each server or the like, reduction in the initial cost and running cost of independently installing and operating the synthetic computer can be expected.

In the above description, a homepage (WWW) server or a telephone server is assumed as the distribution destination. However, if the distribution destination is a broadcasting station, the distributed synthesized voice data is used. It can be broadcast as it is.

[Brief description of the drawings]

FIG. 1 is an illustrative view showing one embodiment of the present invention;

FIG. 2 is a block diagram showing a speech synthesis server of the embodiment in FIG. 1;

FIG. 3 is a functional block diagram illustrating a synthesis computer according to the embodiment in FIG. 2;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 10 ... Speech data distribution system 12 ... Speech synthesis server 14 ... Information source 16 ... WWW (web) server 18 ... Telephone server 26 ... Synthesis computer

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor: Dabin Patrick 2-2-2 Kodai, Seika-cho, Soraku-gun, Kyoto Prefecture Within the International Telecommunications Research Institute, Inc. (72) Yuji Mogi, Seika-cho, Soraku-gun, Kyoto 2-2-2, Kodaidai F-term in International Telecommunication Research Institute, Inc. (reference) 5D045 AA09 AB01 AB26 5K015 GA00 9A001 HH18 JJ25

Claims (5)

[Claims] An audio data distribution system comprising: a receiving unit that receives text data from an information source; a voice synthesizing unit that generates synthesized voice data based on the text data; and a distribution unit that distributes the synthesized voice data to a distribution destination. system. 2. The voice synthesizing means includes a voice waveform signal database and voice unit selecting means for reading voice waveform signal data from the voice waveform database according to index information corresponding to a phoneme string of the text data. 1
The described audio data distribution system. 3. The voice data distribution system according to claim 1, wherein the synthesized voice data synthesized by the voice synthesis means is distributed to a storage system of a computer existing on a network. 4. The audio data distribution system according to claim 1, wherein said distribution destination includes a web server. 5. The audio data distribution system according to claim 1, wherein said distribution destination includes a telephone server.