JP4775236B2 - Speech synthesizer - Google Patents

Description

  The present invention relates to a speech synthesizer that outputs information representing the quality of synthesized speech together with synthesized speech.

Conventionally, an object of the present invention is to provide a new Japanese text synthesizing method using not only phoneme units but also diphone units having the center of the phoneme as a boundary. Further, the present invention provides a Japanese text synthesis method that can synthesize speech with higher naturalness and use a corpus more effectively than a conventional method using only phoneme units. As a technology for the purpose of `` Providing '', in the Japanese text-to-speech synthesis method, in the waveform connection in the chain of vowels and vowels, both the connection at the boundary and the connection at the center of those vowels The speech synthesis unit is selected in consideration of the above. (Patent Document 1).
In the above technique, a scale (cost) that matches the perceptual feature is used when performing phoneme unit selection. As described above, a technique for performing quality evaluation of a synthesized sound by a predetermined cost function or the like is publicly disclosed.
JP2003-208188 (summary, FIG. 2)

In general, the user desires higher quality synthesized speech. However, there are various measures for measuring the quality of synthesized speech, and it is difficult for a general user to judge. Although the quality of synthesized speech can be obtained and presented to the user by the cost function as in the above prior art, the user must consider what quality the calculated value means, and similarly It is difficult to judge whether the quality is good or bad.
Therefore, there has been a demand for a speech synthesizer that allows general users to more intuitively determine the quality of synthesized speech.

A speech synthesizer according to the present invention includes:
An input unit for inputting input text to be converted into speech;
A speech synthesizer that performs speech synthesis according to the content of the input text, and calculates the quality of the synthesized speech for the entire input text using a predetermined arithmetic expression;
A response database storing response messages associated in advance for each quality level of synthesized speech;
A voice response unit that reads out a response message from the response database and outputs the voice;
With
The voice response unit
When the speech synthesizer completes speech synthesis, a response message corresponding to the quality of the speech is read from the response database and output as speech.

  According to the speech synthesizer according to the present invention, since the quality of the synthesized speech is output as a response message, the user can know the quality of the synthesized speech by speech and more intuitively determine whether the quality is good or bad. Can do.

Embodiment 1 FIG.
FIG. 1 is a functional block diagram of speech synthesis apparatus 100 according to Embodiment 1 of the present invention.
The speech synthesizer 100 includes a speech synthesizer 110, a recommendation level selector 120, a speech response unit 130, a speech DB 140, and a response DB 150.
The speech synthesizer 110 receives input text to be read out with synthesized speech, and outputs synthesized speech using data stored in the speech DB 140. In speech synthesis, a corpus-based method is used. Also, the quality of the synthesized speech is calculated by the method described later and output to the recommendation level selection unit 120.
The recommendation level selection unit 120 receives information indicating the quality of the synthesized speech from the speech synthesis unit 110, accesses the response DB 150, and reads a corresponding response message.
The voice response unit 130 outputs the response message read by the recommendation level selection unit 120 as a voice to notify the user of the quality of the synthesized voice by voice.
The voice DB 140 stores data necessary when the voice synthesizer 110 performs voice synthesis.
The response DB 150 will be described later with reference to FIG.

In addition, the speech synthesizer 100 includes an input unit such as a network interface as necessary in order to receive the input text.
The synthesized speech output from the speech synthesizer 110 is assumed to be output in the form of a waveform signal, its sampling data, or the like.

  Note that the “speech synthesizer” in the first embodiment is configured by the speech synthesizer 110 and the recommendation degree selector 120.

The quality of the synthesized speech calculated by the speech synthesizer 110 is a value obtained by evaluating the sound quality using a cost function that associates the physical quantity of speech with perception, which is generated at the stage of generating synthesized speech by the corpus-based speech synthesis method. is there. The cost function is assumed to be predetermined.
Alternatively, the quality of the synthesized speech can be calculated based on the following criteria.
(1) The amount of processing time required to generate synthesized speech.
(2) The input text is divided into mora units and analyzed, and if a mora that deteriorates the sound quality is included, it is evaluated that the sound quality is low.
(3) A feature amount relating to speech speed of the synthesized speech during or after the generation of the synthesized speech. For example, formant transition speed.
(4) A difference value between the mel cepstrum of the synthesized speech after generation and the mel cepstrum estimated for prosody. In addition to the mel cepstrum, the phoneme duration, pitch, LPC coefficient, etc. may be used.

FIG. 2 shows a configuration and data example of a response message table 151 (not shown) stored in the response DB 150.
The response message table 151 includes a “synthesized speech quality” column and a “response message” column.
The “synthesized speech quality” column stores a threshold value representing a quality of synthesized speech calculated by the speech synthesizer 110. In the data example of FIG. 2, a value corresponding to the “cost value” calculated by the speech synthesizer 110 using the above-described cost function is stored, and it is evaluated that a smaller value is a synthesized speech with better quality. To do.
The “response message” column stores response messages corresponding to the quality (= cost value) of synthesized speech generated by the speech synthesis unit 110. In the data example of FIG. 2, for example, if “cost value = 0.10”, the response message is “recommended voice”.

Next, detailed operations of the speech synthesizer 100 will be described step by step.
(1) Input text input The speech synthesizer 110 receives input text to be read out. Note that an input interface is provided in the speech synthesizer 100 as necessary for input. Specifically, for example, direct input by a network interface such as a LAN interface or an operation panel provided on the outer surface of the speech synthesizer 100 can be considered.

(2) Execution / Output of Speech Synthesis The speech synthesizer 110 uses the data necessary for corpus-based speech synthesis, such as prosodic model database, acoustic model database, speech file, etc., stored in the speech DB 140 to input text Generate synthesized speech.
The output format of the synthesized speech may be a data format obtained by sampling a speech waveform, or may be directly output via a speech output device such as a speaker. Alternatively, an electrical signal itself corresponding to the voice waveform may be output.

(3) Calculation of Cost Value When generating the synthesized speech, the speech synthesizer 110 calculates the quality of the synthesized speech using the above cost function and outputs it to the recommendation level selecting unit 120. The output timing may be after the generation of the synthesized speech is completely completed, or may be output sequentially during the synthesis. In the latter case, the recommendation level selection unit 120 may calculate the total cost value.

(4) Selection of Response Message The recommendation level selection unit 120 searches the response message table 151 using the cost value received from the speech synthesis unit 110 as a key. Next, the “response message” column of the corresponding data is read and output to the voice response unit 130.
The data stored in the “response message” column may be only the text of the message, or may store the voice file itself that reads the message.

(5) Output of response message The voice response unit 130 outputs the content of the response message received from the recommendation level selection unit 120 through a speaker or the like.
If the data stored in the “response message” column is only the text of the message, a synthesized speech that reads the text is generated and output. If the message is a voice file itself that reads out the message, the voice file is reproduced and output as a voice.
Note that the voice output timing of the response message may be before the voice synthesizer 110 outputs the synthesized voice or after the output is completely completed. When the output format of the speech synthesizer 110 is waveform sampling data of synthesized speech, the response message may be output as speech along with the output of the data. In this case, the synthesized voice and the voice response are not duplicated and outputted.

  In the first embodiment, the speech synthesizer 110 has been described as performing speech synthesis using a corpus-based method. However, the present invention is not limited to this, and speech synthesis may be performed using a rule synthesis method or a recording / editing method. .

As described above, according to the first embodiment, since the quality of the synthesized speech is output as a response message, the user can know the quality of the synthesized speech by speech, and the quality can be determined more intuitively. Judgment can be made.
Also, obtaining a response message by hearing directly appeals to the user's sensibility rather than simply presenting numerical values, and thus has the effect of increasing the interactivity.

Embodiment 2. FIG.
In the first embodiment, the configuration in which the cost value is calculated using a predetermined cost function based on various parameters obtained when executing speech synthesis has been described.
In the second embodiment of the present invention, a configuration of a speech synthesizer that selects a response message after correcting a cost value when a specific keyword is included in an input text will be described.

FIG. 3 is a functional block diagram of the speech synthesizer 100 according to the second embodiment.
The speech synthesizer 100 according to the second embodiment includes a keyword DB 160. Since other configurations are the same as those in FIG. 1 described in the first embodiment, the same reference numerals are given and description thereof is omitted.

  The keyword DB 160 stores a list of arbitrary keywords stored in a table format, for example. This keyword list is stored in the keyword DB 160 at the time of manufacture by the manufacturer of the speech synthesizer 100 or by setting by the administrator.

Next, the operation of the speech synthesizer 100 according to the second embodiment will be described.
(1) Input text input to (2) Execution / output of speech synthesis Since the operations of these steps are the same as those in the first embodiment, the description thereof is omitted.

(3) Calculation of cost value When the speech synthesizer 110 generates synthesized speech, the speech synthesizer 110 calculates the quality of the synthesized speech using the above-mentioned cost function, and then corrects the calculation result with reference to the keyword DB 160.
As a correction method, a method of correcting the cost value to be lower when the appearance frequency is higher is considered based on how many keywords held in the keyword DB 160 appear in the input text. In addition, a weight can be given to each keyword, and a specific keyword can increase the cost value reduction effect.
The calculated / corrected cost value is output to the recommendation level selection unit 120.

(4) Selection of response message to (5) Output of response message The operations of these steps are the same as those in the first embodiment, and thus description thereof is omitted.

As described above, according to the second embodiment, the calculated cost value can be corrected based on the content of the keyword list. Therefore, depending on the content held in the keyword list, the calculation of the cost value is biased. Can do.
That is, normally, the text input to the speech synthesizer 110 is completely ad hoc, but only when a specific keyword group is input, an appearance as if the quality of the synthesized speech is improved is created. Therefore, the directionality based on the keyword list can be given to the text input to the speech synthesizer 110.
Of course, since it is not preferable to calculate a false cost value from the point of view of integrity, if the input text matches the keyword list, the actual quality of the synthesized speech must be adjusted accordingly. I will.
This function effectively suppresses the input of such words, for example, when the words that are not desirable for output as synthesized speech are input as input text, such as extremely degrading the quality of the synthesized speech. It can be applied to things.

Embodiment 3 FIG.
In the first and second embodiments, the voice response unit 130 outputs a voice response using the content of the response message stored in the response DB 150. This voice response has a special relationship with the synthesized voice. It is assumed to be an inorganic mechanical sound that is generated without any noise.
In the third embodiment of the present invention, the configuration of a speech synthesizer that provides a relationship between synthesized speech and speech response will be described.

FIG. 4 is a functional block diagram of speech synthesis apparatus 100 according to the third embodiment.
The speech synthesizer 100 in FIG. 4 has the same components as in FIG. 1 described in the first embodiment, but the input / output relationship of each part is different from that in FIG. Details will be described in the following operation description.

Next, the detailed operation of the speech synthesizer 100 according to the third embodiment will be described step by step.
(1) Input text input to (3) Cost value calculation Since the operations of these steps are the same as those in the first embodiment, the description thereof is omitted.

(4) Selection of Response Message The recommendation level selection unit 120 searches the response message table 151 using the cost value received from the speech synthesis unit 110 as a key. Next, the “response message” column of the corresponding data is read and output to the speech synthesizer 110.
Note that only the text of the response message is stored in the “response message” column of the response message table 151.

(5) Speech synthesis of response message The speech synthesizer 110 generates synthesized speech that reads the content of the response message received from the recommendation level selector 120. At the time of generation, the same processing as step (2) is performed.
The generated synthesized voice is output to the voice response unit 130.

(6) Output of response message The voice response unit 130 outputs the synthesized voice received from the voice synthesis unit 110 through a speaker or the like.

  As described above, according to the third embodiment, since the response message indicating the quality of the synthesized speech is output in the same speaker or tone as the synthesized speech, the quality of the synthesized speech is more intuitive for the user. An easily understood and interactive speech synthesizer can be provided.

  In the first to third embodiments, a difference may be provided in the voice quality of the response message according to the quality of the synthesized voice. For example, in the third embodiment, when outputting a high-quality synthesized speech with a low cost value, if the quality of the response message is increased or a response message including emotional expressions is output, The interactive feeling given increases accordingly, and the sense of unity between the user and the synthesized speech is enhanced.

  Further, in the first to third embodiments, the number of speakers of the response message is not limited to one speaker, and different speaker response messages may be output for each cost value threshold.

1 is a functional block diagram of a speech synthesizer 100 according to Embodiment 1. FIG. The structure and data example of the response message table 151 stored in the response DB 150 are shown. 6 is a functional block diagram of a speech synthesizer 100 according to Embodiment 2. FIG. 6 is a functional block diagram of a speech synthesizer 100 according to Embodiment 3. FIG.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 100 Voice synthesizer, 110 Voice synthesizer, 120 Recommendation selection part, 130 Voice response part, 140 Voice DB, 150 Response DB, 151 Response message table, 160 Keyword DB.

Claims (4)

An input unit for inputting input text to be converted into speech;
A speech synthesizer that performs speech synthesis according to the content of the input text, and calculates the quality of the synthesized speech for the entire input text using a predetermined arithmetic expression;
A response database storing response messages associated in advance for each quality level of synthesized speech;
A voice response unit that reads out a response message from the response database and outputs the voice;
With
The voice response unit
When the speech synthesizer completes speech synthesis, a response message corresponding to the quality of the speech is read from the response database and output as speech. The response database stores the response messages associated with three or more evaluation ranges for the voice quality level.
The speech synthesizer according to claim 1. A keyword table storing one or more predetermined keywords;
The speech synthesizer
When calculating the quality of synthesized speech, refer to the keyword table,
The speech synthesizer according to claim 1 or 2 , wherein when the content of the input text is included in the keyword table, the quality of the synthesized speech is corrected according to a predetermined rule. The response database stores only the text of the response message;
The speech synthesizer
When the voice response unit voice-outputs the response message, the text of the response message is read and voice synthesis is performed,
The voice response unit
Speech synthesis apparatus according to any one of claims 1 to 3, characterized in that the audio output response message using the voice.

Images