JPH0876782A - Voice synthesizing device - Google Patents

Abstract

PURPOSE: To improve the naturality of synthesized voice by maintaining the specified phoneme continuation time and pause length of single sound and forming phoneme symbol strings and vocal sound and rhythm control parameters. CONSTITUTION: Processing to limit the number of the syllables of one word tone is executed according to the instruction of a speed setting section 10 in a text structure deforming section 8. Next, the phoneme and rhythm control parameter value information is so determined as to extend and contract the phoneme continuation time and pause length of single sound at the ratios according to the speed. This determination is executed by a time length determining section 11 according to the instruction from a speed setting section 10. The phoneme and rhythm symbol strings obtd. from the text structure changed in the pause insertion position by the text structure deforming section 8 and the phoneme and rhythm control parameters formed by the instruction of high-speed utterance by the time length determining section 11 are simultaneously formed and are outputted to a parameter forming section 6. The speed synthesis parameters are formed in parameter formation in such a manner. The parameter systems are sent to a synthesizer to obtain the synthesized voices according to the assigned speed.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesizer for outputting input text as speech according to a certain rule.

[0002]

2. Description of the Related Art FIG. 14 is a block diagram of a conventional speech rule synthesizing device for outputting synthetic speech from text. In FIG. 14, when text is input, the language processing unit 1 derives a word chain from a character string. Therefore, morphological analysis and syntactic analysis are performed. Semantic analysis may also be performed. The phoneme processing unit 4 converts a phoneme sequence forming a word into a phonetic symbol while referring to the environment before and after the word. At the same time, prosodic symbols such as intonation and pause of the whole sentence are generated based on the structural analysis result of the sentence. This phonological / prosodic symbol string is used by the parameter generation unit 6 to generate a time-varying pattern of parameters for generating a speech waveform. The voice waveform generator 7 is driven by this parameter sequence, and finally synthetic voice is output. A rule-synthesized speech corresponding to the text thus input is obtained.

The above-mentioned conventional speech synthesizing device is for sequentially converting input text into speech and outputting it. Generally speaking, speech is synthesized at a normal human speaking speed (about 6.0 to 8.0 mora / sec). To do.

In such speech rule synthesis for converting arbitrary text into speech, by changing the utterance speed of the synthesized speech, for example, if a synthesized speech uttering at a low speed is used, a difficult-to-understand portion or an important portion is clarified. Can be told. On the contrary, by uttering at high speed, it is possible to quickly read unnecessary information portions, so that it is possible to transmit a large amount of information even at the same time. As a result, it is important to change the speed and realize synthetic speech, such as expanding the range of application of speech rule synthesis.

Therefore, conventionally, a method of uttering at high speed by not uttering an unimportant phrase, a method of changing the utterance speed by expanding or contracting the phoneme duration of a single sound or a syllable, and expanding or contracting the utterance time on a segment basis. A method of controlling the speaking rate by controlling the operation has been devised. Control is also performed using a method in which a phoneme that expands and contracts is divided and, for example, only the duration of vowels is changed and the duration of consonant parts is not changed.

[0006] In the conventional method, the phoneme duration time of the part to be expanded / contracted including the pause section is expanded / contracted at a constant rate with reference to the normal vocalization rate. However, when observing the characteristics of speech that humans uttered fast or slowly, the expansion / contraction ratio of the pause length is not constant. There was a problem that it became unnatural.

[0007] In addition, when a person reads a sentence faster,
Some of the pauses up to that point are missing, and the number of mora contained in one phrase tends to increase. It is also known as a well-known fact that when speaking slowly, a pause is inserted and the number of phrases is increased. Therefore, in order to improve the naturalness of the synthesized speech, it is necessary not only to expand or contract the phoneme duration of a single sound and to expand or contract the pause length, but also to change the pause insertion position according to the utterance speed.

[0008]

As described above, in the conventional speech synthesizer, the pause insertion position cannot be changed according to the utterance speed, and the phoneme duration and the pause are simply linear. Since it has been expanded and contracted, there is a drawback in that the synthetic speech becomes unnatural.

An object of the present invention is to provide a speech synthesizer in which the naturalness of synthesized speech is improved by changing the pause insertion position according to the utterance speed and by nonlinearly expanding and contracting the phoneme duration and pause. Especially.

[0010]

SUMMARY OF THE INVENTION According to the present invention, a language processing means for linguistically analyzing input text to create a text structure composed of a chain of tone units, and a speed setting for setting a utterance speed of synthetic speech. Means, text structure transforming means for transforming the text structure created by the language processing means according to the voice speed set by the speed setting means to change the insertion position of the pause, and text obtained by the text structure transforming means. Phoneme processing means for generating a phoneme / prosodic symbol string from the structure, time length determining means for determining the phoneme duration and pause length according to the voice speed set by the speed setting means, and outputting the control parameters, and this time Using the control parameters output from the length determining means, the phoneme / prosodic symbol string from the phoneme processing means is generated as a synthetic speech parameter. And parameter generating means for, is characterized in that a speech waveform generation means for outputting a synthesized speech using a synthesis speech parameters from the parameter generating means.

[0011]

According to the present invention, according to the instruction of the speaking rate, the text structure constituted by the chain of tone units obtained from the result of the linguistic analysis of the text is transformed to set the pause insertion position appropriately, and at the same time, the The phoneme / prosody and control parameter values are determined so that the phoneme duration and pause length are expanded and contracted at different ratios for each speed, and depending on the specified speed designation value, the phoneme duration and pause length of a single note can be determined. The phoneme / prosodic symbol string and the phoneme / prosodic control parameter are generated by keeping the speed constant regardless of the designated speed value. A synthetic voice parameter can be generated based on this information, and a synthetic voice of text can be generated at a desired utterance rate by natural intonation.

[0012]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a block diagram of a voice rule synthesizing apparatus which is an embodiment of the present invention. In FIG. 1, the same parts as those in FIG. 14 are designated by the same reference numerals. 1 differs from the conventional device (FIG. 14) in that a speed setting unit for specifying a speed is provided, and the text structure transformation rule is referred to in accordance with an instruction given from the speed setting unit. The text structure transformation unit that changes the pause insertion position by performing transformation of the above, and when the transformed text structure is output as synthesized speech, expands and contracts the phoneme duration and pause length of a single note at different rates for each speed. As described above, referring to the time length determination unit 11 that determines the phoneme / prosodic control parameter and the specified speed information, if a certain specific utterance speed is given,
There is a point provided in the time length determining unit to keep the phoneme duration and pause length of a single note constant regardless of the specified speed value,
Compared with the conventional device, a speed setting unit 10, a text structure modification unit 8, a rule file 9 having text structure modification rules, and a time length determination unit 11 are added.

The configuration and operation of the speech rule synthesizing apparatus according to the present invention will be described below with reference to FIG. In FIG. 1, an arbitrary text is input to the language analysis unit 1 through some input device (not shown) such as a keyboard, a communication line, a file or a microphone. The language analysis unit 1 analyzes the structure of the input text, determines the structure of the entire text, and analyzes the structure of the analysis rule file 3
Based on the parsing rule stored in, the words included in the text are examined through the dictionary 2 and sentence structure information such as syntactic features and semantic features of each word and dependency relation between each word is determined. When the instruction from the speed setting unit 10 is the normal speed, the text structure including the above information is directly output to the phoneme processing unit 4.

The phonological processing unit 4 examines the words contained in the text structure based on the production rules stored in the production rule file 5, and determines the accent and phonological symbol string of each word. The tone chain is obtained from the text structure, the prosodic symbol string that is the intonation pattern of the entire sentence is determined, and the necessary phonological / prosodic symbol string is generated by the parameter generating unit 6.
When the instruction from the speed setting unit 10 is the normal speed, the phoneme / prosodic symbol string and the prepared phoneme / prosodic control parameter information are output to the parameter generation unit 6. The parameter generation unit 6 creates a voice synthesis parameter and outputs it to the voice waveform generation unit 7. The voice waveform generator 7 is driven by a voice synthesis parameter and outputs the synthesized voice through some output device such as a speaker, a broadcasting device, a communication line or a file. With the above configuration, the synthesized voice at the normal speed can be implemented without any change from the conventional device.

Next, as one embodiment, a specific operation of the present apparatus when the speed setting section gives a high-speed utterance relatively to the normal speed will be described. This is explained using the following input text example sentence.

Example sentence "Let's spread the lunch here and eat today" The text structure analysis of the input text is performed by the language processing unit. In the example, the text structure shown in FIG. 2 is obtained. Here, Head / Modify indicates the syntactic feature of the phrase, and MO indicates the beat of the phrase. NOW indicates the number of words that make up the phrase, (ADJ)
Shows the strength of syntactic connections between phrases.
NO indicates the position of the phrase from the beginning of the sentence, and Phonotype indicates the sequence of syllables. For example, the text "Today is" becomes "Kyowa". phrases indicate syntactic features of a word.

The text structure is grouped into speech units based on each phrase. From the above result, the rest insertion position in normal utterance is shown in a simple notation as follows. Incidentally, "//" visually indicates the position of the pause insertion on the text.

Pause insertion position 1 "Today /// here //
Let's spread the box lunch // Let's eat. "Based on the text structure shown in Fig. 2, the phoneme processing unit obtains a phoneme sequence. This is shown in Fig. 3. For example," Kyo-
The phoneme symbol string of "wa" is "kjo: wa". The numerical value written below the phonetic symbol indicates the accent type of the prosodic word. Such processing is obtained by applying the rule stored in the generation rule file 5.

Then, intonation processing is performed to generate phrase control symbols (P0 to P3) and pause control symbols (S0 to S4). Accent control symbol (DH
~ DHA) is generated, and as a result, the phoneme / control symbol string in FIG. 4 is obtained.

At normal speaking rate, phrase control symbols,
The phoneme / prosodic control parameter value for each control symbol of the pause control symbol, the accent control symbol, and the phoneme duration time (Rate) is set according to a pre-registered specified value (see FIG. 5). This is held in the phoneme processing unit 4.

In the case of normal utterance speed, a parameter sequence for driving the synthesizer is created from the above-mentioned phonological / prosodic symbol string and information on the phonological / prosodic control parameter value, and the result is sent to the synthesizer to generate synthetic speech. can get.

In the case where the utterance speed is changed (increased) In general, when a sentence is uttered quickly, it is not possible to form the tone unit only from the dependency determined from the sentence structure of the text. The number of syllables included in one tone increases and the pause is lost at the same time, compared with the tone unit at the normal speed.
However, when observing human utterances, when the number of syllables in a unit of one tone changes, due to the effect, pauses are uniformly missing throughout the text, and the number of syllables included in one tone is generally large. There is a permissible range, not an equal increase.

Therefore, the text structure modification unit 8 carries out a process of limiting the number of syllables in one tone according to the instruction of the speed setting unit 10. For example, it is assumed that the speed setting unit gives an instruction of a vocalization speed of high vocalization (10.0 mora / second).
The text structure modification unit 8 refers to the table shown in FIG. 6 in the modification unit and adjusts the number of syllables included in one tone according to the designated utterance speed. In the case of this example, the minimum number of syllables in one prosodic phrase is 4, and the maximum is 15.

When a sentence is observed, there is a structure having a syntactic feature that is relatively easy to connect as a prosodic phrase, and conversely, there is a phrase having a syntactic feature that is not connected as a prosodic phrase. For example, a subphrase that indicates the time or place adjacent to the propositional phrase, or an object and other case complements preceding it are collected as a one-tone tone. On the contrary, subordinate phrases and connecting phrases are rarely connected proximately to the preceding and following phrases. Such transformation rules are stored in advance in the text structure transformation rule file 9, and when the number of syllables of one prosodic phrase changes as described above according to an instruction from the speed setting unit 10, the text structure transformation unit 8 changes the text structure Transform using transformation rules.

In the example, the syntactic feature of a word in the text structure is searched, and it is found that there are two prosodic phrases that consist of a subject phrase and a place adverb phrase and are chained, and the total number of syllables of the two prosodic phrases is
Confirm that it is within 16 mora, and combine the place adverbial phrase following the propositional phrase so that it becomes one-tone and change the tone unit. Therefore, the pause between the two episodes of the proposition phrase and the place adverb phrase will be deleted. By such a transformation operation, it becomes possible to obtain a text structure corresponding to a high speaking speed. In the phoneme processing unit 4, this text structure is
By applying the rule of the generation rule file 5, the phoneme sequence corresponding to the text structure is generated. FIG. 7 shows the result.

From the above results, the rest insertion position in normal utterance is shown in a simple notation as follows. It can be seen that the rest insertion position has decreased. Pause insertion position 2 “Today is here // Spread your lunch //
/ Let's eat! ”Then, intonation processing is performed to generate phrase control symbols and pause control symbols. The accent control symbols are generated from the accent information, and the phoneme / control symbol string in FIG. 8 is obtained.

Next, phonological / prosodic control parameter value information is determined so as to expand / contract the phoneme duration and pause length of a single note in proportion to the speed. This is performed by the time length determining unit 11 according to an instruction from the speed setting unit 10. First, the phoneme duration (Rate) of a single note is determined according to the procedure shown in FIG.

[0028] Here, the minimum duration is the minimum duration that can be reasonably heard when voice is synthesized with the duration. On the contrary, the maximum duration is the maximum duration that the sound can be heard without any unnaturalness when the voice is synthesized with the duration. This value is specified according to the performance of the synthesizer. This value can be stored in advance in the time length determination unit 11 or can be given from the outside by a control signal. F1 is a function that returns a return value that does not change at least at a constant ratio in accordance with the speaking speed value (X). For example, it can be given by a function as shown in FIG. A1 to A5 are different constants. The function F1 is not limited to these, and various functions having a non-linear response can be used.

Next, the parameter value (Pn) of the rest length is shown in FIG.
Determine according to the procedure of 11. Here, the minimum pause length is the minimum time length at which the voice can be heard without any unnaturalness when the voice is synthesized with the pause length. On the contrary, the maximum pause length is the maximum length of time that the sound can be heard without any unnaturalness when the voice is synthesized for the length of time. This value is specified according to the performance of the synthesizer. This value can be stored in advance in the time length determination unit 11 or can be given from the outside by a control signal. F2 is a function that returns a return value that does not change at least in a fixed ratio in accordance with the value of the speaking speed specification. In the example, as shown in FIG. 12, it is realized by using linear functions having different range values. B1 to B4 are the coefficients of the function, respectively. The function F2 is not limited to these, and various functions having a non-linear response can be used.

The above-mentioned two procedures determine the phoneme duration and pause, and the information on the phoneme / prosodic control parameter value. As a result, the settings are made as shown in FIG. As shown in FIG. 13, the phoneme duration (Rate) and the pause duration (S1 to S3) are small.

The phoneme / prosodic symbol string obtained from the text structure in which the pause insertion position is changed by the text structure modification unit 8 and the phoneme / prosodic control parameter created by the time length determination unit 11 in response to the high-speed utterance instruction are simultaneously executed. It is created and output to the parameter generation unit. A voice synthesis parameter is created in the parameter generation, and the parameter sequence is sent to the synthesizer to obtain a synthesized voice according to the designated speed.

The method for obtaining an appropriate and natural synthesized speech using the speech synthesizer shown in FIG. 1 has been described above. The text structure modification rules relating to speed control described in the embodiments are not limited to this, and information can be newly added by storing them in the rule file as needed.

Each value in the table for designating the number of syllables for one voice stored in the text structure modification unit 8 can be changed according to the sentence style and the purpose of synthesizing the voice, and is shown in the above-mentioned one embodiment. Not as long. Further, although the Japanese sentence is input to the language analysis unit in the above embodiments, it may be input and processed in word units, or a sentence may be input.

[0034]

According to the present invention, the pause insertion position can be changed by deforming the text structure obtained by linguistically analyzing the input text according to the utterance speed specification.
At the same time, by deciding the phoneme / prosodic control parameter value so that the duration of the single note and the pause length are changed according to the speed without depending on a fixed ratio, the pause insertion position and the duration of the single note and the pause length are changed by movement. Therefore, a natural synthesized voice can be obtained.

[Brief description of drawings]

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

FIG. 2 is a first diagram showing an example of a text structure.

FIG. 3 is a second diagram showing an example of text structure.

FIG. 4 is a first diagram showing examples of phoneme / control symbols.

FIG. 5 shows a first parameter value for each control symbol.
Illustration

FIG. 6 is a diagram showing a designation table for the number of syllables in one-speaking tone

FIG. 7 is a third diagram showing a text structure example.

FIG. 8 is a second diagram showing an example of phoneme / control symbols.

FIG. 9 is a flowchart for determining a phoneme duration.

FIG. 10 is a diagram showing a non-linear function for determining the phoneme duration.

FIG. 11 is a flowchart for determining a rest length.

FIG. 12 is a diagram showing a non-linear function for determining the rest length.

FIG. 13 is a second diagram showing parameter values for each control symbol.

FIG. 14 is a block diagram of a conventional speech synthesizer.

[Explanation of symbols]

1 ... Language processing unit 2 ... Dictionary file 3 ... Analysis rule file 4 ... Phonological processing unit 5 ... Generation rule file 6 ... Parameter generation unit 7 ... Speech waveform generation unit 8 ...
Text structure transformation section 9 ... Text structure transformation rule file 10 ... Speed setting section
11… Time length determination unit

Claims (3)

[Claims] 1. A language processing means for linguistically analyzing input text to create a text structure composed of a chain of word tone units, a speed setting means for setting a utterance speed of synthetic speech, and a speed setting means. A text structure transforming means that transforms the text structure created by the language processing means according to the set voice speed to change the insertion position of the pause, and a phonological / prosodic symbol string from the text structure obtained by the text structure transforming means. The phoneme processing means for generating, the time length determining means for determining the phoneme duration and the pause length according to the voice speed set by the speed setting means, and outputting the control parameters, and the control output by the time length determining means. Parameter generating means for generating a phoneme / prosodic symbol string from the phoneme processing means as a synthetic speech parameter using parameters A voice synthesizing device comprising: a voice waveform generating means for outputting a synthesized voice using the synthesized voice parameter from the parameter generating means. 2. The time length determining means expands / contracts the phoneme duration time and pause length of a single note at different rates for each speed when the set utterance speed is within a predetermined range. The speech synthesizer according to claim 1. 3. The time length determining means sets the phoneme duration and the rest length of a single note to fixed values, respectively, when the set utterance rate is out of a predetermined range. Speech synthesizer.