JPH0675594A - Text voice conversion system - Google Patents

Abstract

PURPOSE:To provide a text voice conversion system which makes good use of each section of the text voice conversion device and facilitates many kinds of voice output methods. CONSTITUTION:The system has plural voice output sections 56a to 56d. An output selection section 55, which selects a voice output section, is provided between a voice synthesis section 54 and the sections 56a to 56d. The text data given to a text analysis section 52 are added with labels which specify a voice output section that outputs analog voice signals corresponding to the text data. The section 52, a synthetic parameter generating section 52 and the section 54 respectively perform the processes designated to themselves against the parts other than the labels, add labels and give the processed data to the next processing section. An output selection section 55 gives the inputted voice waveform data to the voice output section corresponding to the label added to the data.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a text-to-speech conversion system for converting text data into analog voice signals and outputting them.

[0002]

2. Description of the Related Art A text-to-speech conversion system uses arbitrary text data (character character code) as input,
An analog audio signal corresponding to the input is output (for example, a sound output) by adding an accent or intonation.

FIG. 2 shows a hardware configuration of such a text-to-speech conversion system for a Japanese sentence (kanji / kana mixed sentence), and FIG. 3 shows a functional configuration of the text-speech conversion device. It is shown.

In FIG. 2, the text-to-speech conversion system comprises a text data generator (upper device) 10 such as a personal computer, and a text-to-speech converter 20.

The text data generator 10 is, for example,
The text data recorded on the recording medium is given from the data output unit 11 to the text-to-speech converter 20.

In the text-to-speech converter 20, the data input unit 21 takes in the text data provided and temporarily holds it. The control unit 22 includes, for example, a CPU, a ROM, and a RAM, and uses the information stored in the dictionary storage unit 23 to convert the input text data into parameters for voice synthesis (hereinafter, referred to as synthesis parameters). Convert by software. The voice synthesis unit 24 converts the synthesis parameters thus obtained into voice waveform data, and supplies the voice waveform data to the voice output unit 25 including a digital / analog converter or a speaker to output an analog voice signal.

Functionally, the text-to-speech conversion device 20, as shown in FIG. 3, has a text analysis section 31, a word dictionary 32, a synthesis parameter generation section 33, a speech unit storage section 34, and a speech synthesis section. Section 35 (24) and voice output section 36
It consists of (25). That is, the control unit 22 described above
Is a text analysis unit 31 and a synthesis parameter generation unit 33.
Functioning as the word dictionary 3
2 and the voice unit storage unit 34.

In FIG. 3, in the text analysis unit 31,
Using the word dictionary 32, a phoneme and a prosodic symbol string (intermediate language) from the input kanji / kana mixed sentence (text data)
To generate. More specifically, morpheme analysis is performed on a mixed kanji / kana sentence while collating the sentence with the word dictionary 32, and the sentence is divided into morphemes such as words. The word dictionary 32 stores notation of morphemes (words), reading, part of speech, accent position, connection information before and after, and the like. Based on the information stored in the word dictionary 32, the text analysis unit 31 connects the independent word and the independent word or connects the independent word and the adjunct word to create a clause, and A phoneme that describes information such as the reading, accent, intonation, and pose of the input sentence as a character string by connecting the clauses to create a sentence reading and determining the pose position, accent position, strength, etc. in the sentence at that time. , Prosodic symbol string (intermediate language) is generated.

The synthesis parameter generation unit 33 retrieves the voice unit (sound type) data from the voice unit storage unit 34 based on the phoneme and the prosodic symbol string, and sets the duration (sound length) of each phoneme. Then, the pitch pattern (pitch of voice) is determined based on the phoneme and accent information of the prosodic symbol string, and further the amplitude pattern (sound intensity) is determined by the length of each phrase and the like. Generate synthetic parameters. Of these, the voice unit is a basic unit of voice for connecting and creating a synthesized waveform, and various units are prepared in advance according to the type of sound and are stored in the voice unit storage unit 34. .

The voice synthesizing unit 35 generates voice waveform data based on the synthesis parameter, and the voice output unit 36 temporarily holds the voice waveform data in a built-in buffer memory and performs digital / analog conversion according to a clock of a predetermined speed. It is converted into an analog audio signal and output by a built-in speaker, or it is transmitted to another device by a transmission means.

In this way, it is possible to obtain and output an analog voice signal corresponding to the input kanji / kana mixed sentence (text data).

[0012]

By the way, the processing speed of the audio output unit 25 (36) is intended to hear an analog audio signal to a human being, and accordingly, the processing speed becomes accordingly. On the other hand, the processing speed of the control unit 22 functioning as the text analysis unit 31 and the synthesis parameter generation unit 33 and the processing speed of the voice synthesis unit 24 (35) are fast because there is no such restriction. Therefore, a lot of waiting time occurs in the processing of the control unit 22 and the voice synthesizing unit 24, and there is a problem that the operating rate is low and it is not effectively used. Therefore, it is conceivable to provide the control unit 22 and the voice synthesizing unit 24 with a buffer memory for storing processed data one after another to increase the operation rate. However, since the processing speed of the voice output unit 25 is slow, such a buffer It is obvious that the number of writes to memory is larger than the number of calls and the buffer memory overflows, which is not a valid solution.

Further, the conventional text-to-speech conversion system has a problem that it cannot realize various speech output methods because the input system and the output system are one-to-one. For example, you cannot speak multiple documents at the same time,
In addition, it was not possible to broadcast the voice signals related to the same document. In order to achieve such a thing, a plurality of text-to-speech converters 20 are required and the system scale becomes large.

The present invention has been made in consideration of the above points, and the text-to-speech conversion system can utilize each section in the text-to-speech conversion apparatus without waste and can support various voice output methods. Is intended to be provided.

[0015]

In order to solve such a problem, in the present invention, a text analysis unit for generating a phoneme and a prosodic symbol string from input text data using a word dictionary, and a phoneme and a prosodic symbol. A synthesis parameter generation unit that converts the sequence into a voice synthesis parameter using the voice unit storage unit, a voice synthesis unit that converts the voice synthesis parameter into voice waveform data, and a voice waveform data into an analog voice signal. The text-to-speech conversion system provided with a speech output unit for outputting as described above is as follows.

That is, a plurality of voice output units are provided, and an output destination selection unit for selecting a voice output unit is provided between the voice synthesis unit and the plurality of voice output units. In addition, a label defining a voice output unit that outputs an analog voice signal corresponding to the text data is given to the text data to be given to the text analysis unit, and the text analysis unit, the synthesis parameter generation unit, and the voice synthesis unit respectively label the text data. The parts other than are processed by the process assigned to themselves and given a label, and the processed data is given to the next processing unit, and the output destination selection unit gives the input voice waveform data to the data. It is decided to give it to the audio output section corresponding to the label that is set.

Here, the output destination selection unit holds the instructed voice output mode, and the voice output unit which has input the voice waveform data is instructed to the voice output unit corresponding to the label attached to the data. It is preferable to give it at a timing according to the audio output mode.

[0018]

According to the present invention, the text analysis unit uses the word dictionary to generate phonemes and prosodic symbol strings from the input text data, and the synthesis parameter generation unit uses the phoneme and prosodic symbol strings in the speech unit storage unit. And convert it to a voice synthesis parameter,
It is premised on a text-to-speech conversion system in which a speech synthesis unit converts speech synthesis parameters into speech waveform data, and a speech output unit converts speech waveform data into an analog speech signal and outputs it.

Since the voice output unit performs conversion into an analog voice signal by a clock having a speed that takes human auditory characteristics into consideration, the processing speed is slower than that of the text analysis unit, the synthesis parameter generation unit and the voice synthesis unit, and vice versa. For example, the text analysis unit, the synthesis parameter generation unit, and the voice synthesis unit have a margin for processing.

Therefore, in the present invention, a plurality of voice output units are provided to efficiently use the text analysis unit, the synthesis parameter generation unit and the voice synthesis unit. Of course, since a plurality of voice output units are provided, it is necessary to provide an output destination selection unit that selects the voice output unit between the voice synthesis unit and the plurality of voice output units. In order to uniquely specify which text data is given to which voice output section, the text data to be given to the text analysis section should be labeled with a voice output section that outputs an analog voice signal related to this text data. The text analysis unit, the synthesis parameter generation unit, and the speech synthesis unit each perform a process assigned to themselves other than the label to give a label and the processed data to the next processing unit. The output destination selection unit supplies the input voice waveform data to the voice output unit corresponding to the label attached to the data.

Here, the voice output mode is instructed to the output destination selection unit newly provided in the present invention, and the output destination selection unit corresponds the input voice waveform data to the label attached to the data. It is preferable that the sound output section be given at a timing according to the instructed sound output mode so that the outputs from the plurality of sound output sections can be simultaneously or alternately made.

[0022]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described in detail below with reference to the drawings. FIG. 1 shows a text-to-speech conversion system according to this embodiment.

In FIG. 1, the text-to-speech conversion system of this embodiment also comprises a text data generator (upper device) 40 such as a personal computer and a text-speech converter 50. Note that FIG.
Shows the text data generation device 40 in a functional configuration, and the text-to-speech conversion device 50 in a hardware configuration. The functional configuration of the text-to-speech conversion device 50 is the same as the control unit 52 in FIG.
The dictionary storage unit 53 and the dictionary storage unit 53 are divided into a text analysis unit and a synthesis parameter generation unit, and a word dictionary and a speech unit storage unit, respectively.

The text data generator 40 is, for example,
One or more document files (text data) recorded on the recording medium are stored. Here, it is assumed that four document files 41A to 41D are stored.

The text data generator 40 also has a voice output mode setting unit 42. The voice output mode setting unit 42 is a functional entity, and is composed of a keyboard, a CPU, a memory, and the like on the hardware, and the voice output mode instructed by the user, the document file, and the label (specify the voice output unit). Data) combination information. As will be described later, in this embodiment, four audio output units 56a to 56d are used as audio output units.
Is provided.

Here, as the audio output mode, a single file single output mode in which the content of one document file is output from any one audio output unit and a plurality of contents of one document file are output. Single file broadcast output mode to output from the audio output unit, multiple file simultaneous output mode to output the contents of multiple document files simultaneously from separate audio output units, and separate audio output unit to output the contents of multiple document files There are prepared a multiple file alternate output mode for alternately outputting from a plurality of files and a multiple file sequential output mode for sequentially outputting the contents of a plurality of document files from any one audio output unit.

Further, the text data generating device 40 is provided with a file read label adding section 43. When instructed to output the text data, the file read label adding unit 43 extracts the text data from the document file corresponding to the voice output mode set in the voice output mode setting unit 42, and adds the text data to the beginning of the text data. The label for identifying the voice output section associated with the document file relating to the text data is added. The unit amount of the text data to which the label is added may differ depending on the voice output mode. For example, when the audio output mode is the multiple file simultaneous output mode, the file read label adding unit 43 repeats, for example, sequentially extracting a predetermined amount (for example, 512 Bytes) of text data of a plurality of instructed files at that location. Label each quantitation. When the audio output mode is the multiple file alternate output mode, the file read label adding unit 43 alternately takes out the text data of the instructed multiple files one sentence at a time and assigns a label to each sentence.

Here, it is preferable to select the label as a character code string that is unlikely to occur in an actual kanji / kana mixed sentence. Further, although the number of labels to be given to one piece of text data is basically one, there are a plurality of labels in the single file broadcast output mode, for example.

The text data generator 40 also has a data output unit 44. When the voice output mode and the combination information of the document file and the label have been set in the voice output mode setting unit 42, the data output unit 44 outputs the data of the voice output mode and the number of document files to be output as text data. When the output to the voice conversion device 50 is instructed to output the text data, the text data having the label added at the beginning by the file read label addition unit 43 is output to the text voice conversion device 50.

Text-to-speech converter 50 of this embodiment
Is a data input unit 51, a control unit 52 that functions as a text analysis unit and a synthesis parameter generation unit, a dictionary unit 53 that functions as a word dictionary and a speech unit storage unit, a speech synthesis unit 54,
Output destination selection unit 55 and a plurality of audio output units 56a to 56d
It consists of

When the data of the voice output mode and the number of document files to be output is input, the data input unit 51 gives the data to the output destination selection unit 55. Also,
When the text data with the label is input, the data input unit 51 holds the data and holds the data.
I will leave it to the processing by 2.

The control unit 52 includes, for example, a CPU, a ROM and an RA.
M is used to convert the input text data into synthesis parameters for speech synthesis by software while utilizing the information stored in the dictionary storage unit 53. Also adds the label attached to the input text data to the beginning and gives it to the voice synthesis unit 54. The process of converting text data into a synthesis parameter is a text analysis process and a synthesis parameter generation process, which is almost the same as the conventional one, but is slightly different because a label is added to the input text data.

FIG. 4 is a flow chart showing the outline of the text analysis processing according to this embodiment. First, one character data is extracted from the input text data, it is confirmed that a series of text data is not finished, and it is judged whether or not it is the data forming the label (step 100). 102). If it constitutes a label, the process returns to step 100 to take out the next character data. On the other hand, if it is other than the label data, that is, if it is the character data forming the document, the text analysis processing is performed and the process returns to the extraction of the next character data (step 103). In this way, the text analysis process proceeds for character data other than the label, and when the process for the entire series of text data ends, the phoneme / prosodic symbol string obtained by the text analysis process is added to the beginning of the text data. The label is added as it is, and the entire text analysis process is terminated (step 104).

Although the synthesis parameter generation process is not shown in the figure, the process is almost the same from the viewpoint of the label, and the input text is added to the beginning of the synthesis parameter obtained by converting the phoneme / prosodic symbol string. The label that was added to the data is added.

The voice synthesizing unit 54 of this embodiment also performs the synthesizing process on the synthesizing parameters similarly to the conventional voice synthesizing unit to form the voice waveform data. But,
This is different from the conventional voice synthesizer in that a label is added to the beginning of the formed voice waveform data.

Unlike the voice output units 56a to 56d, the control unit 52 and the voice synthesizing unit 54 have no limitation on the processing speed in consideration of human auditory characteristics, and therefore have a high processing speed recognized by the hardware. The data input in step S1 is sequentially processed.

The output destination selection unit 55 has a built-in table shown in FIG. 5 in which the label contents are associated with the plurality of audio output units 56a to 56d. When the label-added voice waveform data is given, the output destination selection unit 55 outputs any one of the voice output data units 56a, 56b, 56c, 56d in which the voice waveform data portion excluding the label is associated with the label.
Give to. At this time, the output destination selection unit 55 outputs the audio waveform data portion to the audio output units 56a to 56d defined by the label by an output method based on the data of the audio output mode and the number of document files to be output.

For example, if the audio output mode is a single file single output mode in which the content of one document file is output from any one audio output unit, the output destination selection unit 55 simply displays the label. The voice waveform data part except for is given to one of the voice output parts associated with the label. If the audio output mode is a single-file broadcast output mode in which the contents of one document file are output from a plurality of audio output units, the output destination selection unit 55 outputs the audio waveform data portion excluding the label. It is given to all the voice output units associated with the assigned labels. further,
When the audio output mode is a multiple file simultaneous output mode in which the contents of a plurality of document files are simultaneously output from different audio output units, the output destination selection unit 55 causes the output destination selection unit 55 to make a cycle of the labels for sequentially applied labeled audio signals. In addition, the voice waveform data portions excluding the labels for the plurality of voice output units are simultaneously given to those voice output units. Furthermore, when the audio output mode is a multiple file alternating output mode in which the contents of a plurality of document files are alternately output from separate audio output units, the output destination selecting unit 55 causes the audio waveform data portion excluding the label to be output. It is sequentially given to one of the voice output units associated with the label. Further, if the audio output mode is a multiple file sequential output mode in which the contents of a plurality of document files are sequentially output from any of the audio output units, the output destination selection unit 55 simply causes the audio waveform data portion excluding the label to be output. Is given to one of the voice output units associated with the label.

The voice output units 56a, 56b, 56c, 5
Each of the 6d has a built-in buffer memory and a digital / analog converter, receives the voice waveform data given from the output destination selection unit 55 in the buffer memory, and outputs the voice waveform data stored in the buffer memory at a predetermined speed. Based on the clock, it is converted into an analog signal by a digital / analog converter and output as a sound from a built-in speaker (not shown) or transmitted to another device by a built-in communication means.

A user who wants to output the contents of one or more desired text files from a desired voice output section by the text-to-speech conversion system having the above configuration sets the voice output mode and outputs the text to be output. It suffices to set the correspondence between the file and the label (voice output section) and activate the output.

In the following, the multiple file simultaneous output mode is set as the audio output mode, the label 01 is set in the document file 41A, and the document file 4 is set as shown in FIG.
The operation will be described assuming that the label 00 is set in 1B and the label 02 is set in the document file 41C.

When the mode or the like is set, it is transmitted from the text data generating device 40 to the text-to-speech converting device 50 that the voice output mode is the multiple file simultaneous output mode and the number of text files related to the output is three. , And then the labeled document file 41
The text data of A, the document file 41B, and the document file 41C are sequentially output to the text-to-speech converter 50.

In the text-to-speech conversion device 50, the multiple file simultaneous output mode and the number of text files 3 are given to the output destination selection unit 55. Thereafter, for the input labeled text data, the control unit 52 uses the dictionary storage unit 53 to perform a text analysis process and a synthesis parameter generation process on a portion other than the label,
The obtained synthesis parameter is delivered to the voice synthesis unit 54 with a label attached. The synthesis parameter portion other than the label is converted into voice waveform data by the voice synthesis unit 54 and given to the output destination selection unit 55 together with the label. In the output destination selection unit 55, the voice output unit 56b in which only the voice waveform data portion of the input voice waveform data with a label is designated by the label based on the preliminarily given plural file simultaneous output mode and the number of text files is 3, 56a,
56c and audio output units 56b, 56a, 56
Document file 41 converted to an analog signal by c
Analog audio signals corresponding to the text data of A, the document file 41B, and the document file 41C are simultaneously output.

Therefore, according to the above embodiment, depending on the voice output mode, the control unit 52 and the voice synthesis unit 54 operate so as to give the voice waveform data to the plurality of voice output units,
The operating ratios of the control unit 52 and the voice synthesizing unit 54 can be increased on average, compared to the conventional case.

Further, when it is attempted to operate a plurality of voice output units, conventionally, it is necessary to provide a plurality of text-to-speech conversion devices or the entire system, but according to this embodiment, the control unit is provided. 52, the dictionary storage unit 53, and the voice synthesizing unit 54 can be commonly used, and the configuration when operating a plurality of voice output units can be made simpler than in the past.

Further, according to the above-mentioned embodiment, the output destination selecting section is provided which has a plurality of audio output sections and selectively controls the supply of the audio waveform data to the audio output sections in the audio output mode. Output method can be performed.
That is, various usage methods can be provided.

In the above embodiment, the voice output mode is set to the text data generating device 40, but it may be set directly to the output destination selecting section 55.

In the above embodiment, various audio output methods are shown. However, a system which can output only the content of one document file from any one audio output unit. May be

Furthermore, the present invention is not limited to the text-to-speech conversion system for Japanese sentences,
It can be naturally applied to a system for converting text data of other languages.

[0050]

As described above, according to the present invention, a plurality of voice output units are provided and an output destination selection unit for selecting a voice output unit is provided between the voice synthesis unit and the plurality of voice output units. The text data provided to the text analysis section,
A label that defines a voice output unit that outputs an analog voice signal corresponding to this text data is given, and the text analysis unit, the synthesis parameter generation unit, and the voice synthesis unit are assigned to each other except for the label. Processing is performed, a label is added, and the processed data is given to the next processing unit, and the output destination selection unit outputs the input voice waveform data to the voice output unit corresponding to the label given to the data. Since it is provided, it is possible to realize a text-to-speech conversion system that can utilize each unit in the text-speech conversion device without waste and can support various voice output methods.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an embodiment.

FIG. 2 is a block diagram showing a conventional hardware configuration.

FIG. 3 is a block diagram showing a conventional functional configuration.

FIG. 4 is a flowchart showing an outline of a text analysis process of the embodiment.

FIG. 5 is an explanatory diagram showing a correspondence relationship between a label and an audio output unit according to the embodiment.

FIG. 6 is an explanatory diagram showing labeled text data according to the embodiment.

[Explanation of symbols]

40 ... Text data generating device, 42 ... Voice output mode setting unit, 43 ... File reading label adding unit, 44 ... Data output unit, 50 ... Text voice converting device, 51 ... Data input unit, 52 ... Control unit (text analysis unit) , Synthesis parameter generation unit), 53 ... dictionary storage unit (word dictionary, speech unit storage unit), 54 ... speech synthesis unit, 55 ... output destination selection unit, 5
6a to 56d ... Voice output unit.

Claims (2)

[Claims] 1. A text analysis unit that generates a phoneme and a prosodic symbol string from input text data using a word dictionary, and a phoneme and a prosodic symbol string as a speech synthesis parameter using a speech unit storage unit. A synthesis parameter generation unit for conversion,
In a text-to-speech conversion system including a voice synthesizing unit for converting a voice synthesizing parameter into voice waveform data and a voice output unit for converting voice waveform data into an analog voice signal and outputting the analog voice signal, a plurality of the voice output units are provided. An output destination selection unit for selecting a voice output unit is provided between the voice synthesis unit and the plurality of voice output units, and an analog voice signal corresponding to the text data is output to the text data given to the text analysis unit. The text analysis unit, the synthesis parameter generation unit, and the voice synthesis unit each perform a process assigned to themselves other than the label to give a label. Then, the processed data is given to the next processing unit, and the output destination selection unit attaches the input voice waveform data to the data. A text-to-speech conversion system characterized by giving to the above-mentioned voice output unit corresponding to a given label. 2. The text-to-speech conversion system according to claim 1, wherein the output destination selection unit holds the instructed voice output mode, and the input voice waveform data is assigned to a label attached to the data. A text-to-speech conversion system, characterized in that it is applied to the corresponding voice output section at a timing corresponding to an instructed voice output mode.