JPH0772889A - Device for forming sound message - Google Patents

Abstract

PURPOSE:To provide a sound message forming device capable of easily forming a sound message provided with a required rhythm characteristic. CONSTITUTION:In a sound message forming device forming the sound message corresponding to an inputted Japanese sentence conforming to a rule, this device is provided with a sound input part 7, a sound analysis/division part 9 analyzing the inputted sound and dividing the input sound according to the sound attribute, a rhythm characteristic extraction part 10 extracting the rhythm characteristic used for revising the rhythm characteristic of a synthetic sound from the divided sound, a speech synthesis information revision part 5 revising synthetic sound information set conforming too the rule according to the extracted rhythm characteristic and a speech synthesis part 6 synthesizing the sound from speech synthesis information.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice message creating apparatus, and more particularly to a voice message creating apparatus capable of easily creating a voice message having a desired prosody characteristic.

[0002]

2. Description of the Related Art A conventional example of this type of speech synthesis method will be described. Conventionally, in services using voice,
Use a pre-determined, already stored voice fragment,
A "speech synthesis method by editing" has been developed in which these are edited to create a voice message. However, this "speech synthesis method by editing" naturally limits the types and contents of services, and also requires a great deal of effort from voice recording to message creation to create a voice message for voice response. , Had the drawback of having to rely on the same announcer. In order to solve these problems, a "rule-based speech synthesis method" for converting Japanese sentences into speech has been proposed. According to this method
There is no limit to the output voice, and the problem of having to secure the same announcer is solved, but since the voices are synthesized based on the rules, intonation, power of unit voice used for voice synthesis, time length etc. It was difficult to arbitrarily modify the prosody characteristics of the speech, and it was difficult to modify the prosody characteristics of the synthesized speech where it sounds unnatural and obtain a voice message having the desired prosody characteristics.

Here, in order to facilitate the modification of the above-mentioned prosody characteristics, the prosody characteristics are displayed in the form of a figure on a display, and a modification operation is visually performed with a mouse or other pointing device to produce a synthesized voice. There is also a proposal to create a voice message while confirming by. However, since this work involves finding auditory unnatural parts and visually modifying the prosody characteristics on the figure, it is a work that requires experience and phonetic knowledge. It was a difficult task for a normal operator. The unnaturalness of the synthesized speech may be perceived, but the correction parameter and the corrected portion may not be clearly pointed out, or a desired intonation may be required depending on the applied service. In such a case, since there is a relationship between the prosody characteristics, it is almost impossible to modify the prosody characteristics on the display.

[0004]

DISCLOSURE OF THE INVENTION The present invention provides a voice message creating apparatus by which an operator who creates a voice message can easily create a voice message having a desired prosody characteristic.

[0005]

[Means for Solving the Problems] In a voice message creating apparatus for creating a voice message corresponding to an input Japanese sentence by a rule, a voice input unit 7 is provided, the input voice is analyzed, and its voice attribute According to the above, a voice analysis / segmentation unit 9 for segmenting the input voice is provided, and a prosody characteristic extraction unit 10 for extracting a prosody characteristic used to correct the prosody characteristic of the synthesized speech from the segmented speech is provided. ,
A voice message creation device comprising a voice synthesis information correction unit 5 for correcting synthesized voice information set by a rule according to the extracted prosody characteristics, and a voice synthesis unit 6 for synthesizing voice from voice synthesis information. did.

[0006]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT An embodiment of the voice message creating apparatus of the present invention will be described with reference to FIG. First, a Japanese text, which is character information including kanji and kana to be converted into a voice message, is input to the Japanese input unit 1. This input information is sent to the voice synthesis information conversion unit 3 via the main control unit 2. The input information sent to the voice synthesis information conversion unit 3 is
Here, punctuation mark information, kana reading string information,
Converted to accent information. Here, the delimiter is a symbol that indicates the degree of connection between clauses, such as strong connection (*), weak connection (/), short pose (-), medium pose (,), long pose (.), And others. There is. The kana reading string is a kana string corresponding to the voice to be uttered, and is usually expressed by a katakana character string. Accent information indicates the position of an accent within a syllable and is represented by a syllable number added from the beginning.

The katakana string is further decomposed into a symbol string representing a unit voice used for voice synthesis, and accent information,
The time length, power, pitch frequency and other prosody characteristics of each unit voice are set according to rules from the delimiter information. The unit voices and prosody characteristics used for voice synthesis are stored in the voice synthesis information storage unit 4 and sent to the voice synthesis information correction unit 5 as necessary. The voice synthesis information correction unit 5 sends the above-mentioned voice synthesis information to the voice synthesis unit 6, and the operator corrects the prosody characteristics of the synthesized voice while listening to the synthesized voice.

The voice message creating apparatus of the present invention uses the prosody characteristics of the voice produced by the operator himself to correct the prosody characteristics of the synthetic speech. Here, it is conceivable to use the operator's own utterance itself as it is, but if this is done, the sense of discomfort with respect to the voice message increases and the serviceability deteriorates. Therefore, only the prosody characteristics of the voice uttered by the operator are focused and used.
This will be described below.

The operator inputs the same voice as the voice message to be created into the voice input section 7 by using a desired prosody. The actual voice input to the voice input unit 7 is stored in the voice signal buffer unit 8 and then voice analyzed by the voice analysis / segmentation unit 9 to be segmented into categories according to the analysis attribute of the voice. The processing in the voice analysis / segmentation unit 9 is the voice segment extraction process and the voice analysis process.
It is divided into two processes. In the voice segment extraction process,
For the audio signal stored in the audio signal buffer unit 8, short-time power is calculated from the beginning of the buffer, and if this power exceeds a threshold value, this is determined to be the beginning of the audio, and similarly from the end of the buffer for a short time. The power is calculated, and when the power exceeds a threshold value, it is determined that the end of the voice, and the voice signal is extracted. In the voice analysis processing, parameters necessary for voice segmentation, such as short-time power, number of zero crossings, analysis residual signal correlation coefficient, cepstrum coefficient difference, and high / low frequency component ratio are calculated at regular time intervals. . This fixed time is called a frame length, and analysis is performed for 30 msec, which is considered to normally maintain a fixed waveform. Such speech analysis is executed for the entire extracted speech section while shifting at regular intervals. This temporal shift for a certain period of time is called a frame shift and is normally set to 5 msec. In the voice segmentation, the input voice is segmented into a plurality of categories by referring to the above-mentioned analysis parameters. When the frame shift is 5 msec, the segmentation accuracy is 5 msec. However, considering speech synthesis based on the extracted prosody characteristics, it is 5 msec.
The accuracy of sec is audibly satisfactory. There are several possible classifications for the segmented categories, from rough classifications to fine classifications, but if the segmented category type and the unit speech type used for speech synthesis match, it is convenient to modify the prosody characteristics. Is good. That is, power, time length and other prosodic characteristics may be extracted for each segmented unit speech, and the prosody characteristics of the corresponding unit speech may be modified. Since speech segmentation is known for input speech, it can fully utilize physical characteristics and a priori knowledge of speech, and compared with speech segmentation technology used in speech recognition for unknown speech. Remarkably high segmentation accuracy is expected.

The prosody characteristic extraction unit 10 performs processing for extracting a prosody characteristic used for modifying synthetic speech from the segmented unit speech. This process is a relatively easy process if the voice segmentation is appropriately performed. That is, the time length of the segmented speech can be easily calculated from the segment boundary, the power information can be easily calculated from the power value of the voice sample within the segment boundary, and the analysis residual for the pitch frequency is also included. It can be calculated from the correlation value. The prosody characteristic for each segmented speech thus obtained is used for modifying the prosody characteristic of the synthesized speech. At this time, it is possible to select the prosody characteristic used for correction. That is, when it is desired to correct only the time length information as a result of the audition of the synthesized voice, only the time length information obtained from the analysis / segmentation of the input voice may be used. The selection of these prosody characteristics to be modified can also be determined while listening to the synthesized voice. When a desired voice is confirmed by a prosody characteristic, a unit voice used for voice synthesis, and other voice synthesis information correction work, and a desired voice is confirmed, the voice synthesis information corresponding to this voice is stored in the voice synthesis information storage unit 4. This completes all operations.

[0011]

As described above, according to the voice message creating apparatus of the present invention, when the "synthesized by rules" sound is unnatural and it is difficult to correct the prosodic characteristics, or When the operator wants to create a synthetic voice with prosodic characteristics that are significantly different from the prosody characteristics set by the rules, the operator uses the prosody characteristics of his or her own vocal sound to synthesize a natural and desired prosody characteristic without changing the sound quality. Sounds can be created easily, and the effort required for voice message creation work can be greatly reduced. Since the voice message creating device of the present invention does not require special phonetic expertise, a general user can easily create a desired voice message.

[Brief description of drawings]

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

[Explanation of symbols]

 1 Japanese input section 2 Main control section 3 Speech synthesis information conversion section 4 Speech synthesis information storage section 5 Speech synthesis information correction section 6 Speech synthesis section 7 Speech input section 8 Speech signal buffer section 9 Speech analysis / segmentation section 10 Prosodic characteristics Extractor

Claims (1)

[Claims] 1. A voice message creating apparatus for creating a voice message corresponding to an input Japanese sentence by a rule, comprising a voice input unit, analyzing the input voice, and inputting the input voice according to the voice attribute. The speech analysis / segmentation unit for segmentation is provided, and the prosody characteristic extraction unit for extracting the prosody characteristic used to correct the prosody characteristic of the synthesized speech from the segmented speech is provided, and according to the extracted prosody characteristic. A voice message creating apparatus comprising: a voice synthesis information correction unit that corrects voice synthesis information set by a rule; and a voice synthesis unit that synthesizes voice from voice synthesis information.