JP2005309164A - Device for encoding data for read-aloud and program for encoding data for read-aloud - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a device for encoding data for read-aloud which can encode a composition mixed with Chinese characters and the Japanese syllabary (text data) by suppressing an amount of data of a dictionary file for decoding and by enabling the dictionary file to be changed without necessitating an exlusive decording device (decoder) at the time of performing decoding, and to rovide a program for encoding data for read-aloud. <P>SOLUTION: The device for encoding data for read-aloud 1 is a device which inputs the composition mixed with Chinese characters and the Japanese syllabary as data for read-aloud and encodes the data into voice streams, and equipped with; a word dictionary accumulating means 7; a morpheme analyzing means 5; a rhythm predicting means 9; a phonemic sequence selection means 11; a stream shaping means 13; and a stream connecting means 15. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a reading data encoding apparatus and a reading data encoding program for encoding a kanji-kana mixed sentence input as reading data into an audio stream.

  In recent years, speech when text data such as kanji mixed text is uttered is reproduced as synthesized speech by using a large amount of speech data (for example, data uttered by various speakers on various sentences). In many cases, attempts have been made to improve the quality of the synthesized speech by using various methods (see, for example, Non-Patent Documents 1 and 2 and Patent Document 1).

For example, the voice of the news read by the announcer is recorded, a voice waveform database is constructed, and voice synthesis is performed using this voice waveform database (a large amount of voice data) (for example, Japanese Patent Application No. 2003-296484 voice). Synthesis method, speech synthesis apparatus and speech synthesis program).
Yuji Matsumoto, Kei Kitauchi, Tatsuo Yamashita, Yoshitaka Hirano, Osamu Ima, Tomoaki Imamura “Japanese Morphological Analysis System“ Chaya ”version 1.5 Instruction Manual”, 1997 “Synthesis of sentence speech based on fundamental frequency pattern generation process model” (The Institute of Electronics, Information and Communication Engineers Journal A Vol. J72-A, No. 1 pp32-40, 1989. 1) JP 2003-271199 A (paragraphs 0029 to 0033, FIG. 1)

  However, all of the speech synthesis techniques that have been attempted in recent years are those that directly convert a kanji-kana mixed sentence into a speech signal (method, apparatus or program), and when the speech signal is transmitted, On the receiving side that receives the signal, there is a problem that a special single device is required to decode the audio signal. That is, when an audio signal is received and played back by a TTS (Text To-speech Synthesis) text stream, the original audio signal is converted and played back unless there is a dedicated decoding device. I could not.

  In addition, when decoding (decoding) a high-quality encoded audio stream on the receiving side (receiving side), the data amount of the decoding dictionary file that must be prepared on the receiving side is large. There is a problem of becoming enormous.

  Furthermore, if the dedicated decoding device is a single device, the dictionary file for decoding is built in and is not supposed to be changed (updated), so it is difficult to change to another dictionary file. There's a problem.

  Therefore, in the present invention, when the above-described problem is solved and decoding (decoding) is performed, a dedicated decoding device (decoding device) is not required, the data amount of the dictionary file for decoding is suppressed, and the dictionary file is It is an object of the present invention to provide a reading data encoding apparatus and a reading data encoding program that can be changed and can encode a kanji-kana mixed sentence (text data).

  In order to solve the above-mentioned problem, the reading data encoding apparatus according to claim 1 inputs a kanji-kana mixed sentence as reading data and encodes the reading data into an audio stream. The phoneme string selection unit, the stream shaping unit, and the stream connection unit are provided.

  According to such a configuration, the reading data encoding apparatus stores a phoneme string by the phoneme string selection unit, and provides a kanji character to the speech encoding apparatus that is an external device that encodes the phoneme string into a speech stream. A phoneme string selection signal for selecting a phoneme string to be encoded is output based on an analysis result obtained by morphological analysis of a kana mixed sentence, a kanji kana mixed sentence, and a phoneme symbol in which the kanji kana mixed sentence is phoneme-notified. A kanji-kana mixed sentence is a Japanese sentence and is so-called text data. These kanji-kana mixed sentences, analysis results, and phoneme symbols are specially used for speech synthesis because a speech stream (connected speech stream) encoded and connected by the device is used for speech synthesis on the receiving side. It can be said that the data has been converted into data. Further, the phoneme string selection means outputs a control signal for controlling how the audio stream encoded by the audio encoding device is connected. This speech coding apparatus encodes a phoneme string into a speech stream (speech coding information; for example, an AAC speech stream) and outputs the speech stream.

  Then, the read-out data encoding apparatus cuts out (extracts) a necessary portion from the audio stream output from the audio encoding apparatus by the stream shaping unit, shapes the audio stream, and the shaped audio stream Are connected based on the control signal output from the phoneme string selection means by the stream connection means, and output as a connected audio stream. The necessary part to be shaped by the stream shaping means is a part used by the stream connecting means, and is determined according to a control signal input to the stream connecting means. That is, when a necessary part is cut out by the stream shaping means, a control signal is input to the stream shaping means directly from the phoneme string selection means or via the stream connection means.

  3. A reading data encoding apparatus according to claim 2, wherein the reading data encoding apparatus inputs a kanji-kana mixed sentence as reading data and encodes the reading data into an audio stream, and stores the word dictionary. Means, morphological analysis means, prosody prediction means, phoneme string selection means, stream shaping means, and stream connection means.

  According to this configuration, the reading data encoding apparatus uses the morphological analysis unit to refer to the word dictionary stored in the word dictionary storage unit to perform morphological analysis of the kanji-kana mixed sentence and to convert the kanji-kana mixed sentence. Output as phonemic symbolic phonemes. The word dictionary includes at least data related to the reading of the word. In addition to the data related to the reading of the word, the word dictionary includes, for example, data related to the word part of speech, the accent type, the connection probability between words, and the connection probability between parts of speech. It is a waste. Subsequently, the reading-out data encoding device uses the prosody prediction means to indicate the prosody of the kanji-kana mixed sentence based on the analysis result obtained by the morphological analysis by the morphological analysis means and the phoneme symbol in which the kanji-kana mixed character is phoneme-notified. Predict the symbol. The prosodic symbols indicate information related to accents and intonations.

  Then, this reading data encoding device stores a phoneme sequence by the phoneme sequence selection means, and transmits the phoneme sequence to a speech encoding device which is an external device that encodes the phoneme sequence into a speech stream. Then, based on the analysis result, the phoneme symbol and the prosodic symbol, a phoneme sequence selection signal for selecting a phoneme sequence to be encoded is output. Further, the phoneme string selection means outputs a control signal for controlling how the audio stream encoded by the audio encoding device is connected. This speech coding apparatus encodes a phoneme string into a speech stream (speech coding information; for example, an AAC speech stream) and outputs the speech stream.

  Then, the read-out data encoding apparatus cuts out (extracts) a necessary portion from the audio stream output from the audio encoding apparatus by the stream shaping unit, shapes the audio stream, and the shaped audio stream Are connected based on the control signal output from the phoneme string selection means by the stream connection means, and output as a connected audio stream.

  4. A reading data encoding apparatus according to claim 3, wherein the reading data encoding apparatus inputs a kanji-kana mixed sentence as reading data and encodes the reading data into an audio stream. The data data storage unit, the phoneme string selection unit, the audio stream output unit, the stream shaping unit, and the stream connection unit are provided.

  According to such a configuration, the data encoding device for reading is based on the phoneme string selection means, the analysis result of the morphological analysis of the kanji-kana mixed sentence, and the phoneme symbol in which the kanji-kana mixed sentence is phoneme-notified. Select a phoneme string stored in the phoneme string encoded data dictionary storage means. The phoneme sequence encoded data dictionary storage means decomposes words constituting a kanji mixed sentence into phonemes, and stores a dictionary related to phoneme sequence encoded data obtained by encoding a phoneme sequence in which the decomposed phonemes are continuous. It is. Further, the phoneme string selection means outputs a control signal for controlling how the phoneme string encoded data (audio stream) is connected.

  Then, the reading data encoding apparatus outputs the phoneme sequence encoded data corresponding to the phoneme sequence selected by the phoneme sequence selection unit as the audio stream by the audio stream output unit, and the necessary data is output from the output audio stream. This part is cut out by the stream shaping means, and the audio stream is shaped.

  Thereafter, the read-out data encoding apparatus connects the audio stream shaped by the stream shaping unit by the stream connecting unit based on the control signal output from the phoneme string selection unit, and outputs the connected audio stream.

  5. A reading data encoding apparatus according to claim 4, wherein the reading data encoding apparatus inputs a kanji-kana mixed sentence as reading data, and encodes the reading data into an audio stream. Means, morphological analysis means, prosody prediction means, phoneme sequence encoded data dictionary storage means, phoneme sequence selection means, speech stream output means, stream shaping means, and stream connection means. .

  According to this configuration, the reading data encoding apparatus uses the morphological analysis unit to refer to the word dictionary stored in the word dictionary storage unit to perform morphological analysis of the kanji-kana mixed sentence and to convert the kanji-kana mixed sentence. Output as phoneme symbols. Subsequently, the reading-out data encoding device uses the prosody indicating the prosody of the kanji-kana mixed sentence based on the analysis result obtained by the morphological analysis by the prosody prediction means and the phoneme symbol in which the kanji-kana mixed sentence is phoneme-notified. Predict the symbol.

  Then, the reading data encoding device selects the phoneme sequence stored in the phoneme sequence encoded data dictionary storage unit based on the kanji kana mixed sentence, the analysis result, the phoneme symbol, and the prosodic symbol by the phoneme sequence selection unit. To do. The phoneme sequence encoded data dictionary storage means decomposes words constituting a kanji mixed sentence into phonemes, and stores a dictionary related to phoneme sequence encoded data obtained by encoding a phoneme sequence in which the decomposed phonemes are continuous. It is. Further, the phoneme string selection means outputs a control signal for controlling how the phoneme string encoded data (audio stream) is connected.

  Then, the reading data encoding apparatus outputs the phoneme sequence encoded data corresponding to the phoneme sequence selected by the phoneme sequence selection unit as the audio stream by the audio stream output unit, and the necessary data is output from the output audio stream. This part is cut out by the stream shaping means, and the audio stream is shaped.

  Thereafter, the read-out data encoding apparatus connects the audio stream shaped by the stream shaping unit by the stream connecting unit based on the control signal output from the phoneme string selection unit, and outputs the connected audio stream.

  The data encoding device for reading according to claim 5 is the data encoding device for reading according to claim 3 or 4, wherein the phoneme sequence encoded data dictionary storage means includes at least a speaker and a speech speed. A plurality of different phoneme sequence encoded data dictionaries are stored, and the audio stream output means includes output switching means for switching and outputting the phoneme sequence encoded data stored in the plurality of phoneme sequence encoded data dictionaries. It is characterized by having.

  According to such a configuration, the reading data encoding apparatus stores a plurality of phoneme sequence encoded data dictionaries in which at least one of the speaker and the speech speed is different in the phoneme sequence encoded data dictionary storage unit, and outputs By switching by the switching means, it is possible to output the most appropriate phoneme sequence encoded data for the kanji mixed sentence.

  The reading data encoding device according to claim 6 is the reading data encoding device according to any one of claims 1 to 5, wherein the stream shaping unit decodes the audio stream. The stream connecting means connects the single units to form the connected audio stream.

  According to such a configuration, the data encoding device for reading can be shaped into the smallest single unit that can be decoded by the stream shaping unit, and the single unit audio stream can be easily connected by the stream connection unit. A connected audio stream to which a single unit audio stream is connected is transmitted and received, and then the connected audio stream is decoded, so that a kanji-kana mixed sentence can be reproduced as an audio signal.

  The data encoding device for reading according to claim 7 is the data encoding device for reading according to any one of claims 1 to 6, wherein the stream connecting means sets a time preset on the time axis. The audio stream is connected with overlapping minutes.

  According to such a configuration, the read-out data encoding apparatus connects the audio streams so that the audio streams overlap each other for a preset time on the time axis by the stream connecting means, and the connected audio streams are connected. It is said. The preset time is, for example, 50% of the length of a single unit of the audio stream.

  The read-out data encoding device according to claim 8 is the read-out data encoding device according to any one of claims 1 to 7, wherein the stream connecting unit connects the audio stream. Quantization accuracy changing means for changing the quantization accuracy in the vicinity of the connection point is provided.

  According to such a configuration, the data encoding device for reading has different quantization accuracy information by changing the quantization accuracy in the vicinity of the connection point when the audio streams are connected by the quantization accuracy changing unit. Audio streams can be connected.

  The read-out data encoding program according to claim 9 inputs a kanji-kana mixed sentence as read-out data, and encodes the read-out data into an audio stream. Means and stream connection means.

  According to such a configuration, the data encoding program for reading is based on the phoneme string selection means, the kana-kana mixed sentence, the analysis result obtained by morphological analysis of the kanji-kana mixed sentence, and the phoneme symbol that represents the kanji-kana mixed sentence as phonemes. Select a phoneme string to be encoded. The read-out data encoding program cuts out a necessary part from the audio stream output from the audio encoding device by the stream shaping means, connects the audio streams to each other by the stream connecting means, and outputs it as a connected audio stream To do.

  According to the first and ninth aspects of the invention, a phoneme sequence to be encoded is selected based on a kanji-kana mixed sentence, an analysis result obtained by morphological analysis of the kanji-kana mixed sentence, and a phoneme symbol in which the kanji-kana mixed sentence is phoneme-notified. To do. Then, a necessary part is cut out from the audio stream output from the audio encoding device, and the audio streams are connected and output as a connected audio stream. For this reason, when decoding a connected audio stream, a dedicated decoding device (decoding device) is not required, and it is possible to reproduce the sound when a kanji-kana mixed sentence is spoken using an existing decoding device. In addition, the kanji mixed kana sentence can be encoded.

  According to the second aspect of the present invention, referring to the word dictionary, based on the analysis result obtained by performing morphological analysis on the input kanji-kana mixed sentence and the phoneme symbol in which the kanji-kana mixed character is phoneme-notified, A prosodic symbol indicating a prosody when a kanji-kana mixed sentence is uttered is predicted, and a phoneme string to be encoded is selected based on the kanji-kana mixed sentence, the analysis result, the phoneme symbol, and the prosodic symbol. Then, a necessary part is cut out from the audio stream output from the audio encoding device, and the audio streams are connected and output as a connected audio stream. For this reason, when decoding a connected audio stream, a dedicated decoding device (decoding device) is not required, and it is possible to reproduce the sound when a kanji-kana mixed sentence is spoken using an existing decoding device. In addition, the kanji mixed kana sentence can be encoded.

  According to the third aspect of the present invention, the phoneme string encoded data dictionary includes a kanji-kana mixed sentence, an analysis result obtained by morphological analysis of the kanji-kana mixed sentence, and a phoneme symbol in which the kanji-kana mixed sentence is phoneme-notified. A phoneme sequence is selected, and phoneme sequence encoded data corresponding to the phoneme sequence is output as an audio stream, and a necessary portion is cut out from the output audio stream to shape the audio stream. Thereafter, the shaped audio stream is connected based on the control signal and output as a connected audio stream. For this reason, when decoding a connected audio stream, a dedicated decoding device (decoding device) is not required, and it is possible to reproduce the sound when a kanji-kana mixed sentence is spoken using an existing decoding device. In addition, the kanji mixed kana sentence can be encoded. Further, by providing the phoneme sequence encoded data dictionary, it is possible to suppress the amount of data in the decoding dictionary file used for decoding and to encode kanji-kana mixed text (text data).

  According to the fourth aspect of the invention, referring to the word dictionary, the kanji-kana mixed sentence is subjected to morphological analysis, and the kana-kana kana based on the result of the morphological analysis and the phoneme symbol in which the kanji-kana mixed sentence is phoneme-notified. Predict prosodic symbols that indicate prosody of mixed sentences. Then, based on the kanji mixed sentence, analysis result, phoneme symbol and prosodic symbol, the phoneme sequence included in the phoneme sequence encoded data dictionary is selected, and the phoneme sequence encoded data corresponding to this phoneme sequence is It outputs as a stream, cuts out a necessary part from the output audio stream, and shapes the audio stream. Thereafter, the shaped audio stream is connected based on the control signal and output as a connected audio stream. For this reason, when decoding a connected audio stream, a dedicated decoding device (decoding device) is not required, and it is possible to reproduce the sound when a kanji-kana mixed sentence is spoken using an existing decoding device. In addition, the kanji mixed kana sentence can be encoded. Further, by providing the phoneme sequence encoded data dictionary, it is possible to suppress the amount of data in the decoding dictionary file used for decoding and to encode kanji-kana mixed text (text data).

  According to the fifth aspect of the present invention, a plurality of phoneme sequence encoded data dictionaries in which at least one of the speaker and the speech speed is different are stored and switched so that the speaker, the speech speed, or both Can be changed, and the phoneme sequence encoded data most suitable for the kanji mixed sentence can be output.

  According to the sixth aspect of the present invention, the audio stream can be shaped into the smallest unit that can be decoded, and the unit can be easily connected.

  According to the seventh aspect of the present invention, the audio stream is overlapped by a predetermined time on the time axis by the stream connecting means, for example, in the case of AAC, time domain aliasing is performed. Cancellation can be performed to reduce aliasing noise at the connection point between audio streams. In addition, by overlapping a plurality of single unit audio streams, the audio signal obtained by decoding the connected audio stream connecting the single unit audio streams is also superimposed, so that sudden changes in the audio signal can be reduced. Can do. Moreover, the quality of the audio signal can be easily improved by alleviating the sudden change in the audio signal.

  According to the eighth aspect of the invention, it is possible to connect audio streams having different quantization information by changing the quantization accuracy in the vicinity of the connection point when the audio streams are connected to each other.

Next, embodiments of the present invention will be described in detail with reference to the drawings as appropriate.
Here, regarding the data encoding device for reading in the two embodiments (the first embodiment and the second embodiment), first, the configuration of the device will be described, and then the operation of the device will be described. I will do it.

<Configuration of Reading Data Encoding Device [First Embodiment]>
FIG. 1 is a block diagram of a reading data encoding apparatus (first embodiment). As shown in FIG. 1, a reading data encoding apparatus 1 inputs text data such as a kanji-kana mixed sentence as reading data and encodes the reading data. , Word dictionary storage means 7, prosody prediction means 9, phoneme string selection means 11, stream shaping means 13, and stream connection means 15.

  In addition, a speech encoding device 3 and a stream distribution device 2 are connected to the reading data encoding device 1. Prior to the description of the configuration of the reading data encoding apparatus 1, these apparatuses 3 and 2 will be described.

  The speech encoding device 3 sequentially encodes the selected phoneme sequence based on the phoneme sequence selection signal output from the reading data encoding device 1 and outputs the selected phoneme sequence to the device 1. 17 and voice encoding means 19.

  The phoneme string storage means 17 is composed of a storage medium such as a hard disk and a control function for controlling input / output of data, and a voice (speech string) uttered by a specific speaker in advance and a morphological analysis of the voice. Speech / analysis data combined with analysis data (including phoneme strings) is accumulated. This speech / analysis data is for creating a speech stream that can be reproduced as a high-quality synthesized speech on the decoding side from a kanji-kana mixed sentence input to the reading data encoding device 1. It can be said that the dictionary data contains various words and various phrases.

  The phoneme string accumulating unit 17 functions based on the phoneme string selection signal output from the reading data encoding apparatus 1, and the control function of the phoneme string accumulating unit 17 functions to generate speech / analysis data including the corresponding phoneme string. Are sequentially output to the voice encoding means 19.

  In this embodiment, the speech of the news read by the announcer described in Japanese Patent Application No. 2003-296484 (speech synthesis method, speech synthesizer, and speech synthesis program) is stored in the phoneme sequence stored by the phoneme sequence storage means 17. The voice waveform database that recorded is used.

  The speech encoding unit 19 sequentially encodes (speech encodes) speech / analysis data output from the phoneme string storage unit 17 and outputs the speech / analysis data to the read-out data encoding device 1 as an audio stream.

  In this embodiment, the speech encoding device 3 is configured separately from the reading data encoding device 1, but may be incorporated in the reading data encoding device 1.

  The stream distribution device 2 distributes the connected audio stream output from the reading data encoding device 1 to a single device or a number of devices on the reception side (decoding side) through a communication line (network) or the like.

  In this embodiment, the audio stream output from the read-out data encoding apparatus 1 is an AAC (Advanced Audio Coding) MPEG stream (hereinafter referred to as AAC audio stream) stream distribution. When the device 2 is connected to an AAC-compatible receiving device (not shown), the stream distribution device 2 and the receiving device are connected using an optical cable, and the AAC audio stream is multiplexed in a band that can be transmitted by the optical cable. (For details, see JEITA CPX-4141 AAC digital interface).

  Further, when the audio stream output from the reading data encoding apparatus 1 is an MP3 audio stream and the stream distribution apparatus 4 is connected to an MP3-compatible receiving apparatus (not shown), a known (existing) method is used. (For example, a method using RTP or the like).

  Incidentally, RTP (Real-time Transport Protocol; Real-time data transfer protocol) is a data communication protocol designed for transferring video signals and audio signals in a format suitable for real-time, Data (video signal, audio signal) is divided into packets in units of time, and data time information is added to the packets and transferred.

Now, each component of the reading data encoding apparatus 1 will be described.
The morpheme analyzing means 5 performs morpheme analysis on the input kana-kana mixed sentence (normal Japanese notation text sentence, so-called text data) with reference to the word dictionary stored in the word dictionary storage means 7. is there. Various methods have been proposed for this morphological analysis. In this embodiment, the method shown in the version 1.5 usage manual of “Japanese morphological analysis system“ tea bowl ”” described in Non-Patent Document 1. Is used.

  The analysis result obtained by the morphological analysis by the morphological analysis means 5 includes at least one piece of information such as the part of speech, accent type, reading, and dependency of each word included in the kanji-kana mixed sentence. It is. Dependency is a kana-kana mixed sentence (general Japanese), where the structure of a sentence is established in the form that a certain phrase is related to (depends on) another phrase. A dependency grammar (dependency grammar) that defines a relationship.

  The morpheme analyzing means 5 generates a phoneme symbol (phoneme symbol string) in which the input kanji-kana mixed sentence is phoneme-notated.

  These analysis results (information on part of speech, accent type, reading, dependency, etc.) and phoneme symbols are output to the prosody prediction means 9 and the phoneme string selection means 11.

  The word dictionary accumulating means 7 is composed of a storage medium such as a hard disk, and accumulates a word dictionary that is referred to when the morpheme analyzing means 5 performs morpheme analysis. The word dictionary includes at least data related to the reading of words used in the kanji-kana mixed sentence input to the reading data encoding device 1, and in this embodiment, the word part of speech, the accent type, Data on the connection probability between words and the connection probability of parts of speech are included.

  “Part of speech of a word” is a noun, a verb, an adverb, an adjective, and the like, and identification information indicating these parts of speech is attached to each word.

  The “accent type” classifies words according to the presence / absence of an accent nucleus and the number of the accent nucleus. Usually, Japanese accents are sensuously divided into two levels: high accents and low accents, and high to low levels immediately after the accented mora (equivalent to kana characters). Migrate to This transition is called the accent nucleus. An n-mora word does not have an accent nucleus, or there are n (n patterns) accent types depending on the position of the accent nucleus (n is a natural number). That is, when the k-mora has an accent, it is called a k-type, and when there is no accent nucleus, it is called a 0-type.

  “Reading” is a romanized representation of how to read a word. The “probability of connection between words” represents the probability of connection between words. “Part-of-speech connection probability” represents the probability of connection between parts of speech.

  The prosody prediction means 9 is based on the analysis result obtained by the morphological analysis by the morphological analysis means 5 and the phoneme symbol string (phonetic symbol string) in which the kanji-kana mixed sentences are phoneme-notated. A prosodic symbol generated by predicting a pattern, pause length (silent time length), intonation and the like is output to the phoneme string selection means 11.

  This prosody prediction means 9 is disclosed in “Synthesis of sentence speech based on a fundamental frequency pattern generation process model” (see IEICE Transactions A Vol. J72-A, No. 1 pp32-40, 1989. 1). The method of predicting prosody (outputting prosodic symbols) is adopted.

  The phoneme string selection unit 11 performs phoneme encoding based on the input kanji-kana mixed sentence, the analysis result and phoneme symbol output from the morpheme analysis unit 5, and the prosody symbol output from the prosody prediction unit 9. A phoneme string selection control signal for selecting a string is output to the phoneme string storage means 17 of the speech encoding device 3.

  The phoneme string selection control signal is used to select, from the phoneme string storage unit 17 of the speech encoding apparatus 3, each of the phoneme strings composed of a plurality of mora into which kanji-kana mixed sentences are divided. For example, if the kanji-kana mixed sentence is “Today is a good weather”, it is divided into “Today”, “Ha”, “Good”, “Weather”, and “Is”, and “Today” (kyou) A phoneme sequence “ha” (ha), a phoneme sequence “good” (ii), a phoneme sequence “tennki”, and a phoneme sequence “de”. Become.

  In addition, the phoneme string selection unit 11 outputs a stream connection control signal (corresponding to a control signal described in claims) for controlling connection between audio streams to the stream connection unit 15.

  The stream connection control signal includes information on an appropriate speech pause length (time), connection time (timing), and an instruction of a connection method, and connection timing and connection of an audio stream connected by the stream connection means 15 It controls the method.

  The connection method instruction is, for example, requantizing the frequency coefficient (DCT component) of the ADTS frame in accordance with the gain G having the larger scale factor value S, and the audio of both windows. For example, when a stream is added after being multiplied by a frequency coefficient of 1/2, Huffman coding is performed, and the stream is converted into an audio stream at a predetermined bit rate, the frame at the connection point of the audio stream to be connected It is to indicate the number and the gain at the connection point.

  For example, when the frequency coefficient for each frame is 256, the information related to the instruction of the connection method is the gain at the connection point of one audio stream A to be connected with Gain A and the connection point of the other audio stream B to be connected. If gain B is given in advance, the stream connecting means 15 simply indicates only the frame number k of the audio stream A and the frame number t of the audio stream B. Can be connected.

  Instead of outputting this stream connection control signal from the phoneme string selection means 11 to the stream connection means 15, a silence phoneme and a phoneme string suitable for the time of this phoneme (time corresponding to the length of speech pause) are selected. It is also possible to output a phoneme sequence selection control signal to the speech encoding device 3.

  The stream shaping unit 13 outputs an audio stream obtained by cutting out and shaping a necessary part from the audio stream output from the audio encoding device 3 to the stream connection unit 15. Note that the necessary part to be shaped by the stream shaping unit 13 is a part used by the stream connection unit 15 and is determined according to the stream connection control signal input to the stream connection unit 15. That is, when a necessary portion is cut out by the stream shaping unit 13, a stream connection control signal is input to the stream shaping unit 13 directly from the phoneme string selection unit 11 or via the stream connection unit 15. Become.

  In this embodiment, the stream shaping means 13 employs 8-bit ADPCM (Adaptive Differential Plus Code Modulation). When the ADPCM output of this audio stream crosses zero and becomes zero, the input audio stream is disconnected and output to the stream connection means 15, whereby special processing is performed by this stream connection means 15. Even without this, it is possible to realize a connection (smooth phoneme connection) between the audio streams without any sense of incongruity.

  ADPCM is a method of compressing the amplitude information of one sample with respect to linear PCM data (audio signal before being streamed), and when the ADPCM output of the audio stream crosses to zero and becomes zero. Indicates the point in time when the output waveform of the ADPCM intersects the time axis (horizontal axis), that is, when the amplitude information becomes zero.

  Further, the stream shaping means 13 shapes the audio stream output from the audio encoding device 3 into the smallest single unit that can be decoded. The smallest unit that can be decoded is a meaningful unit as an audio stream. For example, this meaningful unit is an AAC Audio_Data_Transport_Stream frame (frame) ADTS (ISO / IEC 13818-7: 2003 Chapter 6 Syntax). See) units. Even a single ADTS frame unit alone can be interpreted by a decoding device (decoder, decoding device) or the like.

  The Syncts “1111 1111 1111 1111” is included in the ADTS_fixed_header which is the header part of the ADTS frame. By cutting the audio stream before this syncword, it can be divided into meaningful units as ADTS frames.

  However, in order to connect the audio streams of AAC in the stream connecting means 15, the information such as the profile needs to be the same. On the contrary, if the profile is the same as before, for example, by adding a header part to the data of a part of the ADTS frame stored in the channel_pair_element of the raw_data_block for storing the audio data, a decoder is obtained by adding an header. Interpretation is possible. That is, the stream shaping means 13 shapes the ADTS frame by regarding it as a meaningful unit.

  Based on the stream connection control signal output from the phoneme string selection unit 11, the stream connection unit 15 temporarily converts the audio stream shaped by the stream shaping unit 13 into frequency coefficients in the frequency domain and adds them at the connection point position. Thus, the audio streams are connected to each other, and the quantization accuracy changing means 15a is provided.

  When an appropriate speech pause length (time), connection time (timing), and connection method instruction are input to the stream connection means 15 as a stream connection control signal, ADPCM responds to the length of the speech pause. By connecting the Null data and the audio stream shaped by the stream shaping means 13, connection of the audio stream (connected audio stream) can be executed.

When the audio streams are connected to each other, the quantization accuracy changing unit 15a changes the accuracy of the quantization value when adding the frequency coefficient (DCT coefficient) obtained by converting the audio streams. The frequency coefficient can be expressed by the product of the quantized value and the gain, and the gain is a value having a scale factor value as an exponent. Frequency coefficient K, when the quantization value R, G scale factor value of the gain and S, can be expressed by K = R × G ^S.

  That is, changing the precision of the quantized value means changing the value of the quantized value R by changing (adjusting) the scale factor value S.

  Further, the phoneme sequence selection means 11 outputs a phoneme sequence selection control signal for selecting a silent phoneme and a phoneme sequence that matches the time of this phoneme (a time corresponding to the length of speech pause) to the speech encoding device 3. In such a case, the stream connection control signal is not necessary, and the speech encoding device 3 searches and selects a phoneme sequence in consideration of the time for connecting the speech streams based on the phoneme sequence selection control signal. .

  The stream connection unit 15 is configured to connect the audio streams shaped by the stream shaping unit 13 by overlapping a preset time (set time) on the time axis. In this embodiment, the set time is about half (50%) of one audio stream.

  The output from the stream connection means 15 in this case is shown in FIG. FIG. 5 is a diagram illustrating a method of overlappingly connecting audio streams on the time axis, using an AAC audio stream as an example.

  As shown in FIG. 5, the output from the stream connecting means 15 is obtained by applying a window function (here, a sine window [2048/256 sample]) to an audio signal on the time axis (time domain) to perform DCT transformation and Huffman. Encoding is performed, and the AAC audio stream can be schematically represented in a form in which the audio streams are continuously overlapped by half.

  The connection method between the audio streams in the stream connection means 15 will be described by taking as an example a case where a plurality of phonemes are connected through a 256 sample window (for details on this connection method, refer to “Audio encoding information processing apparatus, audio Encoded information processing program and speech encoded information processing method "(see Japanese Patent Application No. 2004-118361).

  With respect to the frequency coefficient (DCT component) of the ADTS frame at both ends of each phoneme, requantization is performed in accordance with the gain G having the larger scale factor value S, and the audio stream of both windows is, for example, frequency Huffman coding is performed after adding the coefficients by a factor of 1/2, and converted into an audio stream and connected at a predetermined bit rate.

  In this case, the frequency coefficients are each multiplied by ½, and a high-quality connected audio stream can be created by adding and adding gains at an appropriate ratio over several frames. As described above, in the stream connection unit 15, the position of the connection point (near the connection point) for connecting the audio stream and the frequency coefficient to be added are determined by the stream connection control signal output from the phoneme string selection unit 11. It is determined.

  For example, the stream connecting means 15 receives a plurality of audio streams, and processes these audio streams while controlling the quantization accuracy when adding after multiplying the gain G in the frequency domain. Frequency coefficient conversion means (not shown), gain multiplication means (not shown), frequency coefficient addition means (not shown), and speech encoded information conversion means (not shown) are provided.

  The stream connection unit 15 converts the input audio stream into a frequency coefficient by a frequency coefficient conversion unit (not shown), and multiplies the frequency coefficient by a gain G by a frequency coefficient multiplication unit (not shown). . Then, the stream connecting unit 15 adds the frequency coefficient while controlling the quantization accuracy by a frequency coefficient adding unit (not shown).

  Then, the stream connecting unit 15C converts the added frequency coefficient into a quantized value by a quantization unit (not shown) of the speech encoded information converting unit (not shown), and the speech encoded information converting unit (see FIG. A quantization value is coded (encoded) into a Huffman codeword by a noiseless coding unit (not shown), and the bit rate is equal to or lower than a predetermined value by a rate distortion controller unit (not shown). Determine whether.

  Here, when the rate connecting controller 15 (not shown) does not determine that the bit rate is equal to or lower than the predetermined value, the stream connecting unit 15 performs quantization to reduce the bit rate to the predetermined value or lower. The Huffman codeword is converted by controlling the magnitude of the value, and the quantization value is repeatedly controlled until the bit rate after conversion becomes a predetermined value or less. Further, when the stream connection means 15 determines that the bit rate is equal to or lower than a predetermined value by a rate distortion controller section (not shown), the stream connection means 15 (bit stream multiplexer section (not shown)) of the speech coding information conversion means (not shown). (Not shown), the Huffman codewords are rearranged, converted into a streamed audio stream, and output.

Returning to FIG. 1, the description of the configuration of the reading data encoding apparatus 1 will be continued.
According to the reading data encoding apparatus 1, the morphological analysis unit 5 refers to the word dictionary stored in the word dictionary storage unit 7, morphologically analyzes the kanji-kana mixed sentence, and the prosody prediction unit 9 Based on the analysis result obtained by the morpheme analysis by the morpheme analysis unit 5 and the phoneme symbol in which the kanji-kana mixed character is phoneme-notated, a prosodic symbol indicating the prosody of the kanji-kana mixed sentence is predicted. Then, the phoneme sequence selection unit 11 stores the phoneme sequence and encodes the phoneme sequence into an audio stream. The speech encoding device 3 is an external device that mixes kanji characters, analysis results, phonetic symbols, and prosody. Based on the symbol, a phoneme string selection signal for selecting a phoneme string to be encoded is output. Also, the phoneme sequence selection means 11 outputs a stream connection control signal that controls how the audio streams encoded by the audio encoding device 3 are connected. Then, the stream shaping unit 13 cuts out a necessary part from the audio stream output from the audio encoding device 3 to shape the audio stream, and the stream connection unit 15 converts the shaped audio stream into a phoneme. Connection is made based on the stream connection control signal output from the column selection means 11, and the connection audio stream is output. For this reason, when receiving and decoding a connected audio stream, a dedicated decoding device (decoding device) is not required, and the sound when a kanji-kana mixed sentence is spoken using an existing decoding device Can be reproduced so that the kanji-kana mixed sentence can be encoded.

  Further, according to the read-out data encoding apparatus 1, the stream shaping unit 13 shapes the audio stream into the smallest single unit (meaningful unit), so that the stream connecting unit 15 easily converts the single unit. Can be connected to.

  Furthermore, according to the data encoding device for reading out 1, the stream connection unit 15 connects the audio streams so that the audio streams overlap each other for a preset time on the time axis, and the connected audio is connected. It is a stream. For this reason, compared with the case where the audio streams are simply connected continuously, there is no sense of incongruity occurring at the connection location when decoding, and it is possible to prevent deterioration of the audio quality.

  Furthermore, according to the reading data encoding apparatus 1, when the stream connecting unit 15 connects the audio stream, the quantization accuracy near the connection point is changed by the quantization accuracy changing unit 15a. Audio streams with different quantization accuracy can be connected.

  In this data encoding device for reading 1, only kanji-kana mixed sentences are input as reading-out data (read-out text). By inputting to the selection means 11, the morphological analysis means 5, the word dictionary storage means 7, and the prosody prediction means 9 can be omitted.

  In other words, as text-to-speech, text data such as kanji-kana mixed texts and auxiliary information for speech synthesis such as phoneme symbols and prosodic symbols, or text data such as kanji-kana mixed texts are not included. It can be assumed that there is only data specialized for synthesis.

<Operation of Reading Data Encoding Device [First Embodiment]>
Next, operations of the reading data encoding apparatus 1 (first embodiment) (including operations of the audio encoding apparatus 3 and the stream distribution apparatus 2) will be described with reference to the flowchart shown in FIG. (See FIG. 1).
First, the reading data encoding apparatus 1 uses the morpheme analysis unit 5 to analyze the input kanji mixed text (text data) with reference to the word dictionary stored in the word dictionary storage unit 7 ( Step S1), the analyzed analysis result and the phoneme symbol are output to the prosody prediction means 9 and the phoneme string selection means 11.

  Subsequently, the reading data encoding device 1 generates a prosodic symbol by the prosody predicting unit 9 based on the analysis result and the phoneme symbol analyzed by the morphological analyzing unit 5 and outputs the prosodic symbol to the phoneme sequence selecting unit 11 ( Step S2). Then, the reading data encoding device 1 outputs a phoneme sequence selection control signal for selecting a phoneme to be encoded to the phoneme sequence storage unit 17 of the speech encoding device 3 by the phoneme sequence selection unit 11 and A stream connection control signal (control signal) for controlling the connection is output to the stream connection means 15 (step S3).

  Then, the speech encoding device 3 encodes the phoneme sequence output from the phoneme sequence storage unit 17 by the speech encoding unit 19 based on the phoneme sequence selection control signal, and encodes the phoneme sequence. Is output to the reading data encoding apparatus 1 (step S4).

  Then, the read-out data encoding apparatus 1 cuts out a necessary part from the audio stream by the stream shaping unit 13 (step S5), shapes the audio stream, and outputs it to the stream connection unit 15. Then, the read-out data encoding apparatus 1 connects the audio streams output from the stream shaping means 13 by the stream connecting means 15 (step S6), and outputs the connected audio streams to the stream distribution apparatus 2.

  Then, the stream distribution device 2 distributes the connected audio stream via a network (not shown) or the like (step S7).

<Configuration of Reading Data Encoding Device [Second Embodiment]>
FIG. 2 is a block diagram of the reading data encoding apparatus (second embodiment). As shown in FIG. 2, the reading data encoding apparatus 1A inputs text data such as a kanji-kana mixed sentence as reading data, and reads an encoding dictionary to be referred to when the reading data is encoded. The morphological analysis means 5, the word dictionary storage means 7, the prosody prediction means 9, the phoneme string selection means 11A, the stream shaping means 13, the stream connection means 15, the coding dictionary storage means (Phoneme sequence encoded data dictionary storage means) 21 and audio stream output means 23 are provided. In addition, about the structure similar to the structure of the data encoding apparatus 1 for reading shown in FIG. 1, the same code | symbol is attached | subjected and the description is abbreviate | omitted.

  The phoneme string selection unit 11A outputs an encoded phoneme sequence selection control signal for selecting a phoneme sequence (encoded phoneme sequence [phoneme sequence encoded data]) to the audio stream output unit 23, and also mixes kanji characters. Outputs a dictionary switching control signal that specifies the switching of the coding dictionary stored in the coding dictionary storage means 21 and the switching time according to the speech speed at the time of decoding the connected audio stream that is encoded and connected to the sentence. is there. The phoneme string selection means 11 A outputs an audio stream connection control signal to the stream connection means 15, as with the phoneme string selection means 11.

  The encoded phoneme sequence selection control signal includes a kanji-kana mixed sentence input to the apparatus 1A, a phoneme symbol output from the morpheme analysis unit 5, an analysis result and a prosodic symbol output from the prosody prediction unit 9. Is included.

  That is, this phoneme sequence selection means 11A selects the optimal phoneme sequence encoded data from the switched encoding dictionary.

  The dictionary switching control signal includes a speech speed conversion speed parameter for converting the speech speed. The speech speed conversion speed parameter can be arbitrarily input, or a parameter reported in a conventional speech speed conversion research can be used.

  The encoding dictionary storage means 21 is composed of a storage medium such as a large-capacity hard disk, and a plurality of encoding dictionaries (phoneme sequences) in which speech uttered by a plurality of speakers at various speaking speeds is encoded in advance. An encoded data dictionary). In this embodiment, three coding dictionaries (A, B, C) are accumulated. For example, among the three coding dictionaries (A, B, C), the speaking speed of the coding dictionary A is the fastest, then the speaking speed of the coding dictionary B is fast, and the speaking speed of the coding dictionary C Is preset so as to be the slowest.

  The encoded dictionary storage unit 21 outputs phoneme sequence encoded data to the audio stream output unit 23 based on the dictionary switching control signal and the encoded phoneme sequence selection control signal output from the audio stream output unit 23.

  The audio stream output unit 23 receives the dictionary switching control signal and the encoded phoneme sequence selection control signal from the phoneme sequence selection unit 11A, outputs these signals to the encoding dictionary storage unit 21, and the encoding dictionary storage unit 21. Is switched to output to the stream shaping means 13 and includes an output switching means 23a.

  The output switching unit 23a is for switching the phoneme sequence encoded data output from the encoding dictionary accumulating unit 21 in accordance with the switching time included in the dictionary switching control signal, and causing the stream shaping unit 13 to output the data. .

  According to the reading data encoding apparatus 1A, the word dictionary stored in the word dictionary storage unit 7 is referred to by the morphological analysis unit 5, the kana-kana mixed sentence is morphologically analyzed, and the prosody prediction unit 9 Based on the analysis result obtained by the morpheme analysis by the morpheme analysis unit 5 and the phoneme symbol in which the kanji-kana mixed sentence is phoneme-notated, a prosodic symbol indicating the prosody of the kanji-kana mixed sentence is predicted (generated). Then, the phoneme sequence selection means 11A stores the encoded phoneme sequence selection control signal (including kanji kana mixed sentences, analysis results, phoneme symbols and prosodic symbols) and the dictionary switching control signal in the encoding dictionary storage means 21. The phoneme string encoded data stored in the encoded dictionary is selected. Then, the audio stream output unit 23 outputs phoneme sequence encoded data corresponding to the phoneme sequence selected by the phoneme sequence selection unit 11A as an audio stream, and a necessary portion of the output audio stream is stream-shaped. The audio stream is cut out by the means 13 and shaped. Thereafter, the audio stream shaped by the stream shaping unit 13 is connected by the stream connection unit 15 based on the audio stream connection control signal output from the phoneme sequence selection unit 11A, and is output as a connected audio stream. For this reason, when receiving and decoding a connected audio stream, a dedicated decoding device (decoding device) is not required, and the sound when a kanji-kana mixed sentence is spoken using an existing decoding device Can be reproduced so that the kanji-kana mixed sentence can be encoded.

  Further, according to the reading data encoding apparatus 1A, a plurality of encoding dictionaries A, B, and C different in at least one of the speaker and the speech speed are stored in the encoding dictionary storage means 21 and output. By switching by the switching means 23a, it is possible to output the most appropriate phoneme string encoded data for the input kanji mixed sentence.

<Operation of Reading Data Encoding Device [Second Embodiment]>
Next, the operation of the reading data encoding apparatus 1A (second embodiment) will be described with reference to the flowchart shown in FIG. 4 (see FIG. 2 as appropriate).
First, the reading data encoding apparatus 1A uses the morpheme analysis unit 5 to analyze the input kanji mixed text (text data) with reference to the word dictionary stored in the word dictionary storage unit 7 ( In step S11), the analyzed analysis result and the phoneme symbol are output to the prosody prediction means 9 and the phoneme string selection means 11A.

  Subsequently, the reading data encoding apparatus 1A generates a prosodic symbol by the prosody prediction unit 9 based on the analysis result and the phoneme symbol analyzed by the morphological analysis unit 5, and outputs the prosodic symbol to the phoneme string selection unit 11A ( Step S12). Then, the reading data encoding apparatus 1A outputs, to the speech stream output unit 23, the encoded phoneme sequence selection control signal for selecting the phoneme to be encoded and the dictionary switching control signal for switching the encoding dictionary by the phoneme sequence selection unit 11A. At the same time, a stream connection control signal (control signal) for controlling the connection between the audio streams is output to the stream connection means 15 (step S13).

  Then, the reading data encoding apparatus 1A outputs the encoded phoneme sequence selection control signal and the dictionary switching control signal to the encoding dictionary storage unit 21 by the audio stream output unit 23, and the code of the encoding dictionary storage unit 21 The phoneme sequence encoded data output from any of the conversion dictionaries A to C is switched by the output switching unit 23a and output to the stream shaping unit 13 (step S14).

  Then, the reading data encoding apparatus 1A uses the stream shaping unit 13 to cut out a necessary part from the phoneme sequence encoded data (audio stream) (step S15), shapes the audio stream, and sends it to the stream connection unit 15 Output. Then, the read-out data encoding apparatus 1A connects the phoneme sequence encoded data (audio streams) output from the stream shaping unit 13 by the stream connecting unit 15 (step S16), and delivers the stream as a connected audio stream. Output to device 2.

  Then, the stream distribution device 2 distributes the connected audio stream via a network (not shown) or the like (step S17).

  As mentioned above, although embodiment of this invention was described, this invention is not limited to the said embodiment. For example, although the present embodiment has been described as the reading data encoding apparatus 1, the processing of each component of the apparatus 1 (1A) is regarded as a reading data encoding program described in a general-purpose or special computer language. In addition, the processing of each component of the apparatus 1 (1A) may be regarded as a data encoding method for reading that is regarded as each process of generating (encoding) a connected speech stream from a kanji-kana mixed sentence. Is possible. In these cases, an effect similar to that of the reading data encoding apparatus 1 (1A) can be obtained.

1 is a block diagram of a reading data encoding apparatus (first embodiment) according to an embodiment of the present invention. FIG. It is a block diagram of a data encoding device for reading (second embodiment) according to the embodiment of the present invention. It is the flowchart explaining operation | movement of the data encoding apparatus for reading shown in FIG. 1 (1st embodiment). It is the flowchart explaining operation | movement of the data encoding apparatus for reading (2nd embodiment) shown in FIG. It is the figure explaining the method of overlappingly connecting an audio stream on a time axis, taking an AAC audio stream as an example.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1, 1A Data encoding apparatus for reading 2 Stream delivery apparatus 3 Speech encoding apparatus 5 Morphological analysis means 7 Word dictionary storage means 9 Prosody prediction means 11 Phoneme sequence selection means 13 Stream shaping means 15 Stream connection means 15a Quantization accuracy change means 17 phoneme sequence storage means 19 speech encoding means 21 encoding dictionary storage means (phoneme sequence encoded data dictionary storage means)
23 audio stream output means 23a output switching means

Claims (9)

A reading data encoding apparatus for inputting a kanji-kana mixed sentence as reading data and encoding the reading data into an audio stream,
For a speech encoding apparatus that accumulates phoneme sequences and encodes the phoneme sequences into speech streams, the kanji-kana mixed sentence, the analysis result obtained by morphological analysis of the kanji-kana mixed sentence, and the kanji-kana mixed sentence A phoneme string selection means for outputting a phoneme string selection signal for selecting a phoneme string to be encoded on the basis of the phoneme symbol, and for outputting a control signal for controlling connection between the audio streams;
A stream shaping unit that cuts out a necessary part from the audio stream output from the audio encoding device and shapes the audio stream;
Stream connection means for outputting the audio stream cut out by the stream shaping means as a connected audio stream connected based on the control signal;
A data encoding apparatus for reading aloud, comprising: A reading data encoding apparatus for inputting a kanji-kana mixed sentence as reading data and encoding the reading data into an audio stream,
Word dictionary storage means for storing a word dictionary including at least data relating to reading of words included in the kanji-kana mixed sentence;
Referring to the word dictionary stored in the word dictionary storage means, morphologically analyzing the kanji-kana mixed sentence, and outputting a phoneme symbol representing the phonetic representation of the kanji-kana mixed sentence; and
Prosody prediction means for predicting a prosodic symbol indicating the prosody of the kanji-kana mixed sentence based on the analysis result of the morpheme analysis by the morpheme analysis means and the phoneme symbol in which the kanji-kana mixed sentence is phoneme-notified,
A phoneme sequence to be encoded based on the kanji-kana mixed sentence, the analysis result, the phoneme symbol, and the prosodic symbol is stored in a speech encoding apparatus that accumulates the phoneme sequence and encodes the phoneme sequence into a speech stream. A phoneme string selection means for outputting a phoneme string selection signal to be selected and outputting a control signal for controlling connection between the audio streams;
A stream shaping unit that cuts out a necessary part from the audio stream output from the audio encoding device and shapes the audio stream;
Stream connection means for outputting the audio stream cut out by the stream shaping means as a connected audio stream connected based on the control signal;
A data encoding apparatus for reading aloud, comprising: A reading data encoding apparatus for inputting a kanji-kana mixed sentence as reading data and encoding the reading data into an audio stream,
A phoneme sequence encoded data dictionary storage means for storing a phoneme sequence encoded data dictionary that is a dictionary related to phoneme sequence encoded data obtained by encoding the phoneme sequence;
Based on the kana-kana mixed sentence, the analysis result obtained by morphological analysis of the kana-kana mixed sentence, and the phoneme symbol in which the kanji-kana mixed sentence is phoneme-notified, the phoneme string stored in the phoneme string encoded data dictionary storage means A phoneme string selection means for selecting and outputting a control signal for controlling connection between the phoneme string encoded data;
Speech stream output means for outputting phoneme sequence encoded data corresponding to the phoneme sequence selected by the phoneme sequence selection means as an audio stream;
A stream shaping unit that cuts out a necessary part from the audio stream output from the audio stream output unit and shapes the audio stream;
Stream connection means for outputting the audio stream cut out by the stream shaping means as a connected audio stream connected based on the control signal;
A data encoding apparatus for reading aloud, comprising: A reading data encoding apparatus for inputting a kanji-kana mixed sentence as reading data and encoding the reading data into an audio stream,
Word dictionary storage means for storing a word dictionary including at least data relating to reading of words included in the kanji-kana mixed sentence;
Referring to the word dictionary stored in the word dictionary storage means, morphologically analyzing the kanji-kana mixed sentence, and outputting a phoneme symbol representing the phonetic representation of the kanji-kana mixed sentence; and
Prosody prediction means for predicting a prosodic symbol indicating the prosody of the kanji-kana mixed sentence based on the analysis result of the morpheme analysis by the morpheme analysis means and the phoneme symbol in which the kanji-kana mixed sentence is phoneme-notified,
A phoneme sequence encoded data dictionary storage means for storing a phoneme sequence encoded data dictionary that is a dictionary related to phoneme sequence encoded data obtained by encoding the phoneme sequence;
Based on the kanji-kana mixed sentence, the analysis result, the phoneme symbol, and the prosodic symbol, a phoneme sequence stored in the phoneme sequence encoded data dictionary storage unit is selected, and the phoneme sequence encoded data Phoneme string selection means for outputting a control signal for controlling connection;
Speech stream output means for outputting phoneme sequence encoded data corresponding to the phoneme sequence selected by the phoneme sequence selection means as an audio stream;
A stream shaping unit that cuts out a necessary part from the audio stream output from the audio stream output unit and shapes the audio stream;
Stream connection means for outputting the audio stream cut out by the stream shaping means as a connected audio stream connected based on the control signal;
A data encoding apparatus for reading aloud, comprising: The phoneme sequence encoded data dictionary storage means stores a plurality of phoneme sequence encoded data dictionaries in which at least one of a speaker and a speech speed is different,
5. The voice stream output means comprises output switching means for switching and outputting phoneme sequence encoded data stored in the plurality of phoneme sequence encoded data dictionaries. The data encoding device for reading as described. The stream shaping means shapes the audio stream into the smallest unit that can be decoded when decoding,
The reading data encoding apparatus according to any one of claims 1 to 5, wherein the stream connection unit connects the single units to form the connected audio stream.   7. The data encoding for reading according to claim 1, wherein the stream connecting unit connects the audio streams by overlapping a predetermined time on a time axis. apparatus.   The said stream connection means is provided with the quantization accuracy change means which changes the quantization accuracy of the vicinity of a connection point, when connecting the said audio | voice stream. A data encoding device for reading as described in the paragraph. In order to input a kanji-kana mixed sentence as read-out data and encode the read-out data into an audio stream,
For a speech encoding apparatus that accumulates phoneme sequences and encodes the phoneme sequences into speech streams, the kanji-kana mixed sentence, the analysis result obtained by morphological analysis of the kanji-kana mixed sentence, and the kanji-kana mixed sentence A phoneme string selection means for outputting a phoneme string selection signal for selecting a phoneme string to be encoded based on the phoneme symbol, and for outputting a control signal for controlling connection between the audio streams;
A stream shaping unit that cuts out a necessary part from the audio stream output from the audio encoding device and shapes the audio stream;
Stream connection means for outputting the audio stream cut out by the stream shaping means as a connected audio stream connected based on the control signal;
A data encoding program for reading aloud,

Images