JPH11161298A - Speech synthesis method and apparatus - Google Patents

Abstract

(57) [Summary] [PROBLEMS] To provide speech synthesis according to time and case without adding information on volume, utterance speed, and the like to text. A text input by an input unit is converted into a symbol string including a phoneme sequence and prosodic information by language processing in a language processing unit. Synthesis parameter generation unit 1
In step 06, the speech synthesis unit 1 calculates
In step S07, a synthesis parameter used for speech synthesis is generated. At that time, the synthesis parameter generation unit 106 detects the elapsed time from the start of the output of the synthesized sound according to the synthesis parameter corresponding to the input text in the speech synthesis unit 107 based on the measurement result of the utterance time measurement timer 104, and determines the elapsed time. A synthesis parameter is generated such that the characteristics of the synthesized sound are changed in response.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention converts input text data (hereinafter simply referred to as "text") into a symbol string composed of a phoneme sequence and prosodic information by linguistic processing, and generates a synthesized sound from the symbol string. The present invention relates to a speech synthesis method and apparatus, and more particularly, to a speech synthesis method and apparatus for generating a synthesized sound having characteristics according to time or case.

[0002]

2. Description of the Related Art When synthesizing a corresponding voice from a text such as a kanji-kana mixed sentence, in order to make the synthesized sound closer to a natural human utterance, a volume, a utterance speed,
A method as described in JP-A-6-337876 (hereinafter, referred to as the "Japanese Patent Application Laid-Open No. 6-337876") in which information about sound effects is added, and the sound is synthesized by controlling the volume, the utterance speed, and the sound effects using the information. 1 method) has been proposed.

[0003] Japanese Patent Laid-Open No. 6-83381 discloses a technique of synthesizing speech by controlling the utterance speed, pitch, sound quality, and volume using the meanings of idioms and kanji contained in text. A technique (hereinafter, referred to as a second technique) has also been proposed.

[0004]

However, the first method described above has a problem that information must be added to the text, which is complicated. Further, in the first and second techniques described above, the same text is always synthesized at the same volume and utterance speed for the same text. However, when observing the utterance of a human, even when speaking the same sentence, the utterance is changed while changing the characteristics such as the loudness of the voice and the utterance speed in accordance with the time and situation at the time of speaking.

SUMMARY OF THE INVENTION The present invention has been made in view of the above circumstances, and has as its object to add voice synthesis information to a text without adding information relating to volume, utterance speed, and the like. It is an object of the present invention to provide a speech synthesizing method and apparatus which can realize the above.

[0006]

SUMMARY OF THE INVENTION According to the present invention, when a synthesized sound is generated and output from a symbol string comprising a phoneme sequence and prosody information converted and generated by linguistic processing of input text (text data), the input It is characterized in that the elapsed time from the start of the utterance of the synthesized sound corresponding to the text is measured, and the characteristics of the synthesized sound are changed according to the measured elapsed time. Here, at least one of parameters such as the utterance (utterance) speed, pitch, volume, and voice quality may be applied as a parameter (control parameter) for changing the characteristics of the synthesized sound.

In such a configuration, the characteristics of the synthesized sound are automatically changed according to the elapsed time from the start of utterance, so that the listener's attention can be drawn. In particular, it is good to apply the utterance speed as a characteristic of the synthesized speech to be changed, and gradually increase the utterance speed of the synthesized sound according to the elapsed time from the start of the utterance, that is, as the listener gets used to the synthesized sound. . In this case, if the user is unfamiliar with the synthesized sound, the sound will be output in a slow tone, so that the listener can perceive the contents of the synthesized sound correctly by hearing. You don't need to be frustrated because the audio output is in tone. In addition, pitch, volume, voice quality, and the like can be applied to the characteristics of the synthesized speech to be changed, in addition to the utterance speed.

The present invention also provides at least one of various types of timekeeping information including time, date, and day of the week, instead of the above-mentioned elapsed time from the start of utterance, as information serving as a trigger for automatically changing the characteristics of a synthesized sound. It is also characterized in that one is used.

As described above, since the characteristics of the synthesized sound can be changed according to the timekeeping information, it is possible to represent the date and time and personal information for the listener. Further, the present invention holds and manages the number of appearances for each word in all the texts that have been subjected to the output of the synthesized sound, and generates a synthesized sound based on the input text. The frequency of appearance of a word is examined, and the characteristic of the synthesized sound is changed at least between the word that appears for the first time and the word that appears twice or more.

As described above, by changing the characteristics of a synthesized sound between a word that has appeared for the first time and a word that has appeared more than once, the word that has appeared for the first time is output with characteristics that are easy to hear, and the words that have appeared for the second and subsequent times are output. The word thus output can be output with characteristics that do not cause annoyance to the listener.

Further, according to the present invention, if there is a request to output a synthesized voice for the next text before the output of the synthesized voice corresponding to the input text is completed, the text portion corresponding to the unoutput synthesized voice is deleted. It is characterized by being simplified, and generating and outputting the synthesized speech of the simplified text part instead of the original text part. That is, according to the present invention, the next text input before the end of the audio output of the text currently being processed is used as a trigger to change the way of reading the unoutputted text portion and terminate the audio output of the text portion in a short time. Features.

As described above, when the output of the next text is requested before the output of the synthesized speech corresponding to the input text currently being subjected to the speech output processing is completed, the output of the synthesized text corresponding to the unoutputted synthesized speech is handled. The text portion to be output is simplified, and a synthesized speech of the simplified text portion (for example, a summary) is output. , And can be synchronized in time. In addition,
Instead of simplifying only the text part corresponding to the unoutput synthesized voice, the text part corresponding to the unoutput synthesized voice is combined with at least the first sentence of the next text, and the combined text is simplified. You may do it.

Further, if the output of the next text is requested before the output of the synthesized speech corresponding to the input text currently subjected to the speech output processing is completed, the utterance speed of the unoutput synthesized speech is reduced. By controlling to speed up, the voice output of the text portion that has not been output can be ended in a short time and the voice output of the next requested content can be immediately shifted, so that the time can be synchronized.

Further, if the output of the next text is requested before the output of the synthesized speech corresponding to the input text which is the target of the current speech output processing is completed, the volume of the unoutputted synthesized speech is gradually reduced. Even if it is switched to the output of the synthesized sound for the next text when the volume reaches a predetermined level, the voice output of the unoutputted text portion is terminated in a short time, and Can be immediately shifted to the audio output of the requested content, so that time synchronization can be achieved.

[0015]

Embodiments of the present invention will be described below with reference to the drawings. [First Embodiment] FIG. 1 is a block diagram showing a schematic configuration of a speech synthesizer according to a first embodiment of the present invention.

In FIG. 1, an input unit 101 is for inputting a text to be subjected to speech synthesis. For example, a large-capacity storage device such as a keyboard, an OCR (optical character reading device), a floppy disk device, a magnetic disk device, and the like. , A communication interface (which controls input of text transmitted via a communication medium).

The linguistic processing unit 103 performs a linguistic analysis on the text of a sentence mixed with, for example, kanji and kana input from the input unit 101, and a word dictionary in which reading and accent types are registered in addition to grammatical information such as part of speech information. The processing is performed using 102. That is, the language processing unit 103 collates the input text with the word dictionary 102 to obtain accent type, reading, and part of speech information of words and phrases included in the text, and obtains an accent type associated with the part of speech information, By determining the boundary, the sentence is converted into a reading form of a sentence mixed with kanji and kana, and a prosody information (a phonetic symbol string composed of a phoneme series and an accent type) is generated. This language processing unit 103
The phonological sequence and prosodic information generated by the above are passed to the synthesis parameter generating unit 106.

Upon receiving the phoneme sequence generated by the language processing unit 103, the synthesis parameter generation unit 106 refers to the speech unit file 105. The speech unit file 105 stores a large number of speech units created in advance. The speech unit analyzes a voice uttered by an announcer or the like and obtains a predetermined feature parameter of the voice, and then, in a predetermined synthesis unit, for example, a Japanese syllable unit, all syllables included in the Japanese voice are described above. It is created by cutting out from the feature parameters.

At this time, the synthesis parameter generation unit 106
For example, utterance (utterance) for determining the characteristics of the synthesized speech
Various parameters such as speed, pitch, volume, and voice quality are controlled independently, which will be described later.

The speech synthesis unit 107 generates a sound source and performs digital filtering processing on the phoneme parameter and the prosody parameter generated by the synthesis parameter generation unit 106 and a volume parameter described later (for the input text) to produce a synthesized speech. And the speaker 108
Output to When the output of the synthesized speech corresponding to the input text is started (at the start of utterance), the speech synthesis unit 107 activates a timer for measuring time (hereinafter, referred to as a utterance time measurement timer) 104 to produce the utterance. The elapsed time from the start is measured. Then, at the end of the output of the synthesized speech corresponding to the input text, the speech synthesis unit 107 stops the time measurement by the timer 104 and clears the measured value.

The synthesis parameter generation unit 106
Refers to the measurement value of the utterance time measurement timer 104 when generating the phoneme parameter and the prosody parameter corresponding to the input text, and refers to the measurement value, that is, the synthesized speech (for the input text) by the speech synthesis unit 107. In accordance with the elapsed time from the start of the output (at the start of utterance), various parameters such as the utterance speed, pitch and voice quality are independently controlled and reflected on the phonological parameter and the prosodic parameter (the pitch is different from the prosodic parameter). Only). Further, the synthesis parameter generation unit 106 generates a parameter of a volume corresponding to the elapsed time, and sets the parameter in association with the phoneme parameter and the prosody parameter.

More specifically, as the time elapses from the start of the utterance, the utterance speed is increased stepwise (up to a certain predetermined speed as an upper limit). That is, in the present embodiment, at the initial stage when the output of the synthesized speech corresponding to the input text is started, the listener utters at a relatively slow speed on the assumption that the listener is not used to the synthesized speech,
As the listener gets used to the synthesized sound over time, the utterance speed is increased stepwise (gradually). In this way, at a stage where the user is not accustomed to the synthesized sound, the utterance is performed in a relatively slow tone, so that the listener can perceive correctly even if they are not accustomed to the synthesized sound. Since the utterance is made quickly, no annoyance is required. Also, the utterance speed (characteristic) of the synthesized sound
Is variable, so that the listener's attention can be drawn.

The parameters other than the utterance speed, such as the pitch, volume, and voice quality, are varied, for example, at random over time. This can prevent the listener from getting tired of the synthesized sound, and can further draw the listener's attention.

The above-described method of controlling the parameters and the types of the parameters are merely examples, and the present invention is not limited thereto. In short, by controlling at least one of various parameters such as the utterance speed, pitch, volume, and voice quality according to the elapsed time from the start of utterance, it is only necessary to change the characteristics of the synthesized speech. Attention can be drawn.

The technique itself for controlling the parameters such as the utterance speed, pitch, volume, and voice quality to change the characteristics of the synthesized speech is described in the above-mentioned Japanese Patent Application Laid-Open Nos. Hei 6-337876 and Hei 6-83381. The description is omitted here because it is well known. [Second Embodiment] FIG. 2 is a block diagram showing a schematic configuration of a speech synthesizer according to a second embodiment of the present invention, and the same parts as those in FIG.

The difference between the voice synthesizer of FIG. 2 and the voice synthesizer of FIG. 1 is that the voice synthesizer of FIG. 1 includes a timer 204 having a calendar function and a clock function instead of the utterance time measuring timer 104 of FIG. Instead of the synthesized parameter generation unit 106, at least one of the parameters such as the utterance speed, pitch, volume, and voice quality is controlled based on the clock data (calendar data, time data) of the clock unit 204. In addition to the provision of the synthesis parameter generation unit 206 for changing the characteristics, a condition table 209 in which various conditions for changing the characteristics of the synthesized voice are registered is newly provided.

In the configuration shown in FIG. 2, when the text of a kanji-kana sentence, such as news or mail, to be subjected to speech synthesis is input from the input unit 101, the language processing unit 10
3 collates the input text with the word dictionary 102, obtains the accent type, reading, and part-of-speech information on the words and seasons included in the text, and determines the accent type and boundary associated with the part-of-speech information, The kanji kana-mixed sentence is converted into a reading format to generate a phonemic sequence and prosodic information.

The synthesis parameter generation unit 206 sequentially reads out corresponding speech units from the speech unit file 105 according to the phoneme sequence generated by the language processing unit 103,
The phoneme parameters corresponding to the phoneme sequence are generated by connecting the read speech units. Further, the synthesis parameter generation unit 106 generates a prosody parameter including a pitch pattern and voiced / unvoiced information based on the prosody information generated by the language processing unit 103.

At this time, the synthesis parameter generation unit 206
For example, time information and date at the time of utterance (speech) as time (time measurement) data measured by the timekeeping unit 204,
Condition table 209 based on calendar information such as days of the week
See The condition table 209 includes, for example, a speech rate to be applied for each time range when 24 hours a day is divided into a morning time range, a day time range, an evening time range, and a night time range. The values of parameters such as pitch, volume, and voice quality are registered for each day of the week (or dates classified as holidays and non-holidays). Also, the condition table 209
Registers the values of parameters such as the utterance speed, pitch, volume, and voice quality to be applied to a predetermined specific date such as a birthday.

The synthesis parameter generation unit 206 refers to the condition table 209 based on the time information of the utterance time indicated by the timer unit 204 and the calendar information such as the date and the day of the week. , The values of parameters such as the utterance speed, pitch, volume, and voice quality corresponding to the day of the week (or a date classified as a holiday or a non-holiday). When the date of the utterance coincides with a specific date (such as a birthday) registered in the table 209, the synthesis parameter generation unit 206 determines the time of the utterance (the time range to which the utterance belongs) and the day of the week. , The values of parameters such as the utterance speed, pitch, volume, and voice quality corresponding to the specific date are extracted. Then, the synthesis parameter generation unit 206
Generates a phonemic parameter and a prosodic parameter corresponding to the input text while controlling the corresponding parameter with the parameter value extracted from the condition table 209. The synthesis parameter generation unit 206 sets the parameter values extracted from the condition table 209 in association with the generated phoneme parameters and prosody parameters.

The speech synthesis unit 107 generates a sound source and performs digital filtering processing on the phoneme parameter, the prosody parameter, and the volume parameter generated by the synthesis parameter generation unit 206 to generate a synthesized speech (for the input text). , To the speaker 108.

As described above, in the present embodiment, the characteristics of the synthesized speech can be changed according to the time range and the day of the week. Thus, for example, in the morning of a weekday, the utterance speed = slow, the pitch = high (treble), the volume = loud, and the voice quality =
Improve awakening by using female sound, and utterance speed at night =
By setting it to be slow, pitch = low (low temperature), volume = low, and voice quality = male sound, it is possible to provide comfort. Also, on certain dates, such as the listener ’s birthday,
It is also possible to set the characteristics of the synthesized speech suitable for that.
In other words, by changing the characteristics of the synthesized speech, the date and time and personal information for the listener can be represented. [Third Embodiment] FIG. 3 is a block diagram showing a schematic configuration of a speech synthesizer according to a third embodiment of the present invention, and the same parts as those in FIG.

The difference between the speech synthesizer of FIG. 3 and the speech synthesizer of FIG. 1 is that the speech synthesizer shown in FIG. 1) It includes an appearance word management unit 304 for holding and managing the number of appearances of a word, and based on the number of appearances of the word managed by the appearance word management unit 304 instead of the synthesis parameter generation unit 106 in FIG. , A synthesis parameter generation unit 306 that controls at least one of parameters such as a utterance speed, a pitch, a volume, and a voice quality to vary characteristics of a synthesized voice. The language processing unit 303 converts the information of each word included in the text input from the input unit 101 into the appearing word management unit 3
04.

In the configuration shown in FIG. 3, when a text of a sentence mixed with Chinese characters, such as news or mail, to be subjected to speech synthesis is input from the input unit 101, the language processing unit 30
3 collates the input text with the word dictionary 102, obtains the accent type, reading, and part-of-speech information on the words and seasons included in the text, and determines the accent type and boundary associated with the part-of-speech information, The kanji kana-mixed sentence is converted into a reading format to generate a phonemic sequence and prosodic information. At the same time, the language processing unit 303 notifies the word storage management unit 304 of the information on the words included in the text. Appearance word storage management section 304 includes language processing section 303
When the word information notification is received from, the number of appearances of the corresponding word held and managed by itself is incremented by one.

The synthesis parameter generation unit 306 sequentially reads out corresponding speech units from the speech unit file 105 according to the phoneme sequence generated by the language processing unit 303,
The phoneme parameters corresponding to the phoneme sequence are generated by connecting the read speech units. The synthesis parameter generation unit 306 generates a prosody parameter including a pitch pattern and voiced / unvoiced information based on the prosody information generated by the language processing unit 303.

At this time, the synthesis parameter generation unit 306 refers to the appearance word storage management unit 304, and the number of appearances (held and managed by the appearance word storage management unit 304) for each word to be subjected to speech synthesis. , While independently controlling various parameters such as utterance speed, pitch, volume, and voice quality, generate phonemic parameters and prosodic parameters corresponding to the input text. The synthesis parameter generation unit 306 generates a volume parameter of a value corresponding to the number of appearances for each word, and sets the volume in association with the corresponding phoneme parameter and prosody parameter.

The speech synthesizer 107 generates a sound source from the phoneme parameter and the prosody parameter generated by the synthesis parameter generator 306 and performs digital filtering processing to generate a synthesized voice (with respect to the input text). Output to

As described above, in the present embodiment, it is possible to change the characteristics of the synthesized speech based on the number of appearances of each word to be subjected to speech synthesis, which is held and managed by the appearance word storage management unit 304. it can. Thus, for example, a word that appears for the first time is uttered with a slow tone, so that the listener can correctly perceive even an unfamiliar (unfamiliar) word, and a word that appears twice or more utters at a standard speed. That saves the listener from getting frustrated. Note that a configuration may be adopted in which the utterance speed of a word having a higher number of appearances can be increased (up to a predetermined speed). In addition, a word that appears for the first time uses a parameter value different from the standard, and a word that appears twice or more applies a predetermined standard parameter value. Speaking at a different rate, pitch, volume, or vocal quality can also draw the listener's attention. [Fourth Embodiment] FIG. 4 is a block diagram showing a schematic configuration of a speech synthesizer according to a third embodiment of the present invention, and the same parts as those in FIG.

The difference between the speech synthesizer shown in FIG. 4 and the speech synthesizer shown in FIG. 1 is that the speech synthesis unit 107 shown in FIG. 1 is replaced by a phoneme parameter and a prosodic parameter used for speech synthesis. A speech synthesizing unit 407 having a buffer (output waiting buffer) 407a for holding together with the unuttered text) is provided, and a sentence conversion unit 404 is provided instead of the speech time measuring timer 104 in FIG. . When the utterance for the next text is required before the utterance based on the phoneme parameter and the prosody parameter in the buffer 407a is completed, the sentence conversion unit 404 responds to the user's selection designation in the buffer 407a. It has a function of creating a summary based on an unuttered word string. The sentence conversion unit 404 also causes the voice synthesis unit 407 to rewrite the unvoiced phonological parameters and prosodic parameters in the buffer 407a so as to increase the utterance speed according to the user's selection designation, or to change the unvoiced word string. It also has a function (control function) of giving an instruction to fade out the synthesized sound to the voice synthesis unit 407. The sentence conversion unit 404 further has a function of combining the unspoken word string in the buffer 407a with the next text (sentence) according to the user's selection designation to simplify it.

The difference between the speech synthesizer of FIG. 4 and the speech synthesizer of FIG. 1 is that, instead of the input unit 101 in FIG.
An input unit 401 for notifying a new utterance request (for the next text) to the sentence conversion unit 404 is also provided. The input unit 401 corresponds to the text input by the input unit 401 instead of the language processing unit 103 in FIG. A linguistic processing unit 403 for generating a phonological sequence and prosodic information, and a phonological sequence and prosodic information corresponding to the abstract given from the sentence converting unit 404, respectively.
Then, instead of the synthesis parameter generation unit 106 in FIG.
It has a synthesis parameter generation unit 406 that generates corresponding phonological parameters and prosody parameters according to the phonological sequence and prosody information generated by the language processing unit 403.

In the configuration of FIG. 4, when an utterance request is given from the input unit 401 and a text of a kanji-kana mixed sentence to be subjected to speech synthesis is input, the language processing unit 403 executes the input text and the word dictionary. By comparing 102 with the words and seasons included in the text, the accent type, reading, and part of speech information are obtained, and the accent type and the boundary associated with the part of speech information are determined, so that the kanji kana mixed sentence can be read. A conversion is performed to generate a phoneme sequence and prosody information.

The synthesis parameter generation unit 406 sequentially reads out corresponding speech units from the speech unit file 105 in accordance with the phoneme sequence generated by the language processing unit 403,
The phoneme parameters corresponding to the phoneme sequence are generated by connecting the read speech units. Further, the synthesis parameter generation unit 406 generates a prosody parameter including a pitch pattern and voiced / unvoiced information based on the prosody information generated by the language processing unit 403.

The speech synthesis unit 407 receives the phoneme parameters and the prosody parameters generated by the synthesis parameter generation unit 406, and temporarily stores them in the buffer 407a. Then, the speech synthesis unit 407 performs a sound source generation and a digital filtering process on the basis of the phonemic parameters and the prosodic parameters in the buffer 407a in conjunction with the utterance speed of the synthesized speech to generate a synthesized speech for the input text. Then, the signal is output to the speaker 108.

As described above, it is assumed that the next utterance request is input from the input unit 401 during the output of the synthesized voice for the input text, that is, before the utterance for the parameters in the buffer 407a ends. .

In this case, the sentence conversion unit 404 makes an inquiry to the speech synthesis unit 407 to investigate a word string (in the input text) corresponding to the unvoiced parameter in the buffer 407a (ie, an unvoiced word string). Then, summarization processing for creating a summary based on the unuttered word string, or sentence combining for simplifying the unspoken word string by combining it with a text (sentence) from the input unit 401 for which a new utterance request has been made Simplification processing, processing for giving an instruction to the speech synthesizer 407 to increase the utterance speed of the unuttered word string, or giving an instruction to the speech synthesizer 407 to fade out the synthesized sound of the unuttered word string Execute any one of the processes. Here, which one to execute can be selected and designated by a user operation in advance.

First, the case where the digest creation processing is specified will be described. In this case, the sentence conversion unit 404 uses, for example, “Funasaka et al.,“ Summary of related newspaper article by redundancy reduction ”, IEICE technical report, from the unuttered word string held in the buffer 407a of the speech synthesis unit 407. NLC96-15, pp.39-
46, 1996-07 ", and summarization is performed by a summarization technique. Then, the sentence conversion unit 404 sends the created summary to the language processing unit 403, and requests language processing. At the same time, the sentence conversion unit 404 invalidates, for example, clears the phoneme parameter and the prosody parameter corresponding to the unvoiced word string in the buffer 407a in the speech synthesis unit 407.

When the language processing unit 403 receives the digest from the sentence conversion unit 404, it performs linguistic processing on the summary before performing linguistic processing on the new text input from the input unit 401, and performs the corresponding phonological sequence and prosodic information. Is generated and output to the synthesis parameter generation unit 406. Subsequently, the language processing unit 403 starts language processing on the new text.

Upon receiving the summary phoneme sequence and the prosody information from the language processing unit 403, the synthesis parameter generation unit 406 generates a corresponding phoneme parameter based on the phoneme sequence, and similarly generates a corresponding prosody based on the prosody information. Generate parameters.

The speech synthesizer 407 receives the summary phoneme parameters and prosody parameters generated by the synthesis parameter generator 406, and temporarily stores them in the buffer 407a. Generation and digital filtering are performed to generate a synthesized speech for the summary, and output to the speaker 108. Accordingly, if there is a next output request before the currently output voice ends, the content currently being output can be completed in a short time, and the process can immediately shift to the process for the next utterance request.

Instead of creating a summary as described above, a predetermined keyword is simply extracted from an unuttered word string, and a synthesized sound of the extracted keyword (or keyword string) is generated. It does not matter. For that purpose, a word that is a candidate for a keyword may be registered in a table or the like in advance, and a word that matches the registered word may be searched from an unuttered word string. The creation of an abstract from an unuttered word string and the extraction of a keyword are similar in that the unuttered word string (a sentence) is simplified.

Next, a case where the text combining / simplifying process is specified will be described. In this case, the sentence conversion unit 40
Reference numeral 4 combines an unuttered word string held in the buffer 407 of the speech synthesis unit 407 with, for example, the first sentence in a new text input from the input unit 401 to create a summary thereof. . Then, the sentence conversion unit 404 passes the created summary and the subsequent character string in the new text to the language processing unit 403, and requests language processing. At the same time, the sentence conversion unit 404 sends a buffer 407 to the speech synthesis unit 407.
The phonetic parameter and the prosodic parameter corresponding to the unvoiced word string in a are cleared (invalidated). It should be noted that the unuttered word string and the entire new text may be combined to create a summary.

When the language processing unit 403 receives the digest and the subsequent character string in the new text from the sentence conversion unit 404, the language processing unit 403 performs linguistic processing on the received sentence to generate a corresponding phoneme sequence and prosodic information. , To the synthesis parameter generation unit 406.

Upon receiving the phoneme sequence and the prosody information from the language processing unit 403, the synthesis parameter generation unit 406 generates a corresponding phoneme parameter based on the phoneme sequence, and similarly generates a corresponding prosody parameter based on the prosody information. Generate.

The speech synthesis unit 407 receives the phoneme parameters and the prosody parameters generated by the synthesis parameter generation unit 406 and temporarily stores them in the buffer 407a, and based on the phoneme parameters and the prosody parameters,
A sound source is generated and a digital filtering process is performed to generate a synthesized speech, which is output to the speaker 108. Accordingly, if there is a next output request before the currently output voice ends, the content currently being output can be completed in a short time, and the process can immediately shift to the process for the next utterance request.

Next, a case in which an increase in the utterance speed (increase in utterance speed) is specified will be described. In this case, the sentence conversion unit 404 gives an instruction to the speech synthesis unit 407 to increase the utterance speed of the unuttered word string,
It causes the language processing unit 402 to start language processing on the new text.

Upon receiving the above instruction from the sentence conversion unit 404, the speech synthesis unit 407 subjects the utterance speed to at least one of the phonemic parameter and the prosodic parameter corresponding to the unvoiced word string held in the buffer 407a. Make changes to be faster. And the speech synthesis unit 407
Generates a synthesized voice by performing sound source generation and digital filtering processing based on the changed phonemic parameters and prosodic parameters, and outputs the synthesized voice to the speaker 108.
Accordingly, if there is a next output request before the currently output voice ends, the content currently being output can be completed in a short time, and the process can immediately shift to the process for the next utterance request.

Next, the case where fade-out is designated will be described. In this case, the sentence conversion unit 404
An instruction to fade out the utterance of the unuttered word string is given to the speech synthesizer 407 and the language processor 402
Starts language processing for the new text.

Upon receiving the above instruction from the sentence conversion unit 404, the speech synthesis unit 407 generates a sound source based on the phoneme parameters and the prosody parameters corresponding to the unvoiced word string held in the buffer 407a. A synthetic speech is generated by performing a digital filtering process. On this occasion,
The voice synthesizer 407 outputs the synthesized voice to the speaker 108 while gradually lowering the volume of the synthesized voice so that the volume becomes equal to or lower than a predetermined level after a predetermined time T. And the speech synthesis unit 407
When the volume of the synthesized voice falls below a predetermined level, the output of the synthesized voice is forcibly terminated, and the unvoiced word string and the corresponding phonemic parameters and prosodic parameters in the buffer 407a are cleared (invalidated). Thus, if there is a next output request before the end of the currently output voice, the content currently being output can be forcibly terminated in a short time, and the process can immediately proceed to the process for the next utterance request. In addition, since the volume is gradually lowered and the forced termination is performed when the volume becomes equal to or lower than the predetermined level, it is possible to proceed to the output processing of the next request content without giving a sense of incongruity to the listener.

As described above, in the present embodiment, if there is a next output request before the currently output voice ends, the content being output currently ends in a short time without giving the listener an uncomfortable feeling. For example, in the car navigation device, before the voice output of the guidance of the left or right turn at an intersection is completed, the guidance of the guidance of the left or right turn at the next intersection may be performed. It is suitable when a voice output request is issued.

In the fourth embodiment, any one of the process of summarizing, the process of combining and simplifying sentences, the process of giving an instruction to increase the utterance speed, and the process of giving an instruction to fade out are performed by a user operation. Although described as being selectable, it may be fixedly determined that only one of the above processes is performed.

Each of the above-described speech synthesizers having the structures shown in FIGS. 1 to 4 is a computer, for example, FIG.
A program for a sentence-to-speech synthesis process (sentence-to-speech conversion process) for generating speech from text, as applied in the first to fourth embodiments, is recorded in a personal computer 501 having a built-in speaker 108 as shown in FIG. The recording medium, for example, the CD-ROM 502,
It can also be realized by reading and executing the program recorded in the ROM 502 by the personal computer 501. Further, the program for the sentence / speech synthesis processing can be supplied from a recording medium such as a floppy disk, a memory card, or a communication medium such as a network in addition to the CD-ROM 502.

[0062]

As described in detail above, according to the present invention, the characteristics of the synthesized speech can be automatically changed according to the elapsed time from the start of the utterance even for the same sentence. Further, according to the present invention, even for the same sentence, the characteristics of the synthesized sound can be automatically changed according to time information such as the time of occurrence and calendar information.

Further, according to the present invention, even for the same word, the characteristics of synthesized speech can be automatically changed according to the number of appearances. Further, according to the present invention, when the next output request is issued before the currently output voice ends, the content currently being output is completed in a short time without giving a listener a sense of incongruity, and the next output request is issued. It is possible to shift to processing promptly.
As described above, according to the present invention, how to read the same sentence can be automatically changed according to time and case.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a schematic configuration of a speech synthesizer according to a first embodiment of the present invention.

FIG. 2 is a block diagram showing a schematic configuration of a speech synthesizer according to a second embodiment of the present invention.

FIG. 3 is a block diagram showing a schematic configuration of a speech synthesizer according to a third embodiment of the present invention.

FIG. 4 is a block diagram showing a schematic configuration of a speech synthesizer according to a fourth embodiment of the present invention.

FIG. 5 is an exemplary external view of a personal computer capable of realizing the speech synthesizer shown in FIGS. 1 to 4;

[Explanation of Codes] 101, 401: Input unit 102: Word dictionary 103, 303, 403: Language processing unit 104: Speech time measurement timer 105: Speech unit files 106, 206, 306, 406: Synthesis parameter generation unit 107, 407 voice synthesis unit 108 speaker 204 clock unit 209 condition table 304 appearance word management unit 404 text conversion unit (control means) 407a buffer

Claims (16)

[Claims] 1. A speech synthesis method for converting input text data into a symbol string composed of a phoneme sequence and prosody information by language processing and generating a synthesized speech from the symbol string. A voice synthesizing method characterized by measuring an elapsed time from the start of utterance and changing characteristics of the synthesized sound according to the measured elapsed time. 2. A speech synthesis method for converting input text data into a symbol string comprising a phoneme sequence and prosodic information by language processing and generating a synthesized sound from the symbol string. A voice synthesizing method characterized by acquiring at least one piece of timekeeping information and changing a characteristic of the synthesized sound according to the acquired timekeeping information. 3. A speech synthesis method for converting input text data into a symbol string comprising a phoneme sequence and prosodic information by language processing, and generating a synthesized sound from the symbol string, wherein all of the output of the synthesized sound are performed. In the case where the number of appearances is retained and managed for each word in the text data of and the synthesized speech is generated based on the input text data, the number of appearances is examined for each word in the text data. A speech synthesis method characterized by changing characteristics of a synthesized sound between a word that appears and a word that appears two or more times. 4. A speech synthesis method for converting input text data into a symbol string composed of a phoneme sequence and prosodic information by language processing and generating a synthesized speech from the symbol string. If there is a request to output a synthesized voice for the next text data before output is completed, the text data portion corresponding to the unoutput synthesized voice is simplified, and the synthesized voice of the simplified text data portion is output. A speech synthesis method characterized by generating and outputting in place of the original text data portion. 5. A speech synthesis method for converting input text data into a symbol string composed of a phoneme sequence and prosodic information by language processing and generating a synthesized speech from the symbol string. If there is a request to output a synthesized voice for the next text data before output is completed, the text data portion corresponding to the unoutput synthesized voice and at least the first sentence of the next text data are combined, and A speech synthesis method characterized by simplifying a combined sentence portion, and generating and outputting a synthesized speech of the simplified sentence portion instead of the original sentence portion. 6. A speech synthesis method for converting input text data into a symbol string composed of a phoneme sequence and prosodic information by language processing and generating a synthesized speech from the symbol string. A speech synthesizing method characterized in that when there is a request to output a synthesized voice for the next text data before the output is completed, control is performed so as to increase the utterance speed of a non-output synthesized voice. 7. A speech synthesis method for converting input text data into a symbol string composed of a phoneme sequence and prosody information by language processing and generating a synthesized speech from the symbol string. If there is a request to output a synthesized voice for the next text data before the output is completed, the unoutput synthesized voice is output while gradually lowering the volume. A speech synthesis method characterized by switching to output of a synthesized speech for the text data. 8. A speech synthesizer for converting input text data into a symbol string composed of a phoneme sequence and prosodic information by language processing and generating a synthesized speech from the symbol string. Utterance time measurement means for measuring an elapsed time from the start of utterance synthesis parameter generation means for generating a synthesis parameter to be used for speech synthesis from a phoneme sequence corresponding to the text data and prosody information, wherein the utterance time measurement means Synthetic parameter generating means for generating the synthetic parameters so that the characteristics of the synthetic sound are varied according to the measured elapsed time, and voice synthesizing means for generating synthetic sounds in accordance with the synthetic parameters generated by the synthetic parameter generating means A speech synthesizer comprising: 9. A speech synthesizer for converting input text data into a symbol string comprising a phoneme sequence and prosodic information by language processing and generating a synthesized speech from the symbol string. Clocking means for acquiring at least one of the clocking information; and synthesizing parameter generating means for generating a synthesizing parameter to be used for speech synthesis from a phonological sequence corresponding to the text data and prosodic information. A synthesis parameter generation unit that generates the synthesis parameter such that a characteristic of the synthesis sound is changed according to timing information at the time of utterance of the synthesis sound corresponding to the text data; and a synthesis parameter generation unit that generates the synthesis parameter. A voice synthesizing means for generating a synthesized voice in accordance with the synthesis parameter. 10. A speech synthesizer for converting input text data into a symbol string composed of a phoneme sequence and prosodic information by language processing and generating a synthesized sound from the symbol string, wherein all of the output of the synthesized sound are performed. An appearance word management unit that holds and manages the number of appearances for each word in the text data; and a synthesis parameter generation unit that generates a synthesis parameter to be used for speech synthesis from a phoneme sequence and prosody information corresponding to the text data. For each word in the text data, the number of appearances is checked based on the management content of the appearing word management means, and the characteristics of the synthesized sound change at least between the word that first appears and the word that appears twice or more. Parameter generation means for generating the synthesis parameter as described above, and the synthesis parameter generated by the synthesis parameter generation means. And a voice synthesizing means for generating a synthesized voice in accordance with data. 11. A language processing means for converting input text data into a symbol string comprising a phoneme sequence and prosody information by language processing, and a phoneme sequence and prosody information corresponding to the text data converted by the language processing means. Synthesizing parameter generating means for generating a synthesizing parameter to be used for speech synthesis from a synthesizing parameter; buffer means for holding the synthesizing parameter generated by the synthesizing parameter generating means and text data corresponding to the parameter; A voice synthesizer for generating and outputting a synthesized voice in accordance with the stored synthesis parameters; and a request for outputting a synthesized voice for the next text data before the generation and output of the synthesized voice by the voice synthesizer are completed. The text data portion in the buffer means corresponding to the unoutput synthesized sound. Simplified, the conjunction to perform language processing by the language processing unit, a voice synthesizing apparatus characterized by comprising a text converting means for invalidating the corresponding contents held in the buffer means. 12. A language processing means for converting input text data into a symbol string comprising a phonological sequence and prosodic information by linguistic processing, and a phonological sequence and prosodic information corresponding to said text data converted by said linguistic processing means. Synthesizing parameter generating means for generating a synthesizing parameter to be used for speech synthesis from a synthesizing parameter; buffer means for holding the synthesizing parameter generated by the synthesizing parameter generating means and text data corresponding to the parameter; A voice synthesizer for generating and outputting a synthesized voice in accordance with the stored synthesis parameters; and a request for outputting a synthesized voice for the next text data before the generation and output of the synthesized voice by the voice synthesizer are completed. The text data portion in the buffer means corresponding to the unoutput synthesized sound. A sentence that combines at least the first sentence of the next text data, simplifies the combined sentence portion, performs language processing by the language processing unit, and invalidates the corresponding held content in the buffer unit. A speech synthesizing device comprising: a conversion unit. 13. A speech synthesizer for converting input text data into a symbol sequence comprising a phoneme sequence and prosodic information by language processing and generating a synthesized speech from the symbol sequence. Synthesis parameter generation means for generating synthesis parameters to be used for speech synthesis from prosody information; buffer means for holding the synthesis parameters generated by the synthesis parameter generation means and text data corresponding to the parameters; and the buffer Means for generating and outputting a synthesized sound in accordance with the synthesis parameters held in the means; and a request for outputting a synthesized sound for the next text data before the generation and output of the synthesized sound by the voice synthesis means are completed. If there is, the synthesis parameter in the buffer means corresponding to the unoutput synthesized sound Speech synthesis apparatus characterized by comprising a control means for performing control for rewriting as utterance speed is quickened. 14. A speech synthesizer for converting input text data into a symbol sequence comprising a phoneme sequence and prosodic information by language processing, and generating a synthesized speech from the symbol sequence. Synthesis parameter generation means for generating synthesis parameters to be used for speech synthesis from prosody information; buffer means for holding the synthesis parameters generated by the synthesis parameter generation means and text data corresponding to the parameters; and the buffer Speech synthesis means for generating and outputting a synthesized sound according to the synthesis parameters held in the means,
If there is a request to output a synthesized sound for the next text data before the generation and output of the synthesized sound is completed, the volume gradually decreases according to the synthesis parameters in the buffer means corresponding to the unoutputted synthesized sound. Sound synthesis means for generating and outputting a synthesized sound so as to generate and output a synthesized sound according to a synthesis parameter corresponding to the next text data when the sound volume falls below a predetermined level. A speech synthesizer characterized by the following. 15. Input text data is converted into a symbol string comprising a phoneme sequence and prosodic information by language processing,
A program for generating a synthesized sound from the symbol string, wherein the program monitors an elapsed time from the start of utterance of the synthesized sound corresponding to the text data, and varies a characteristic of the synthesized sound according to the elapsed time. A computer-readable recording medium on which is recorded. 16. Input text data is converted into a symbol string comprising a phoneme sequence and prosodic information by language processing,
A program for generating a synthetic sound from the symbol string, wherein if a synthetic sound output request for the next text data is requested before output of the synthetic sound corresponding to the text data is completed, A computer-readable recording medium storing a program for simplifying a text data portion corresponding to a synthesized voice of the above, and generating and outputting the synthesized voice of the simplified text data portion in place of the original text data portion.