JP4297433B2 - Speech synthesis method and apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a speech synthesizing method for providing higher quality synthesized speech than the conventional one in response to the case that synthesized speech is reproduced, and also to provide its system. <P>SOLUTION: The speech synthesizing method comprises steps of: retrieving an elementary speech unit used for synthesis at each unit in an input text by a DB retrieval part 2 from a database 1 storing a speech corpus including the elementary speech units of various units configuring a language; selecting combination of the optimum elementary speech unit from among the elementary speech units at each unit in the retrieved text by an elementary speech unit selector 3; determining by a state determining part 4 whether audio equipment reproducing the synthesized speech is a head phone or a speaker; outputting each elementary speech unit to a connection part 6 as it is with the head phone, or routing each elementary speech unit through a signal processor 5 with the speaker, and outputting it to the connection part 6 after signal processing correcting a rhythm to the elementary speech unit is performed. <P>COPYRIGHT: (C)2006,JPO&amp;NCIPI

Description

  The present invention converts text input directly from a keyboard or the like, or read and input from a storage medium that stores text in advance, or input from another device or the like via a communication medium, into sound and reproduced (output). The present invention relates to text-to-speech synthesis technology.

  Currently, text-to-speech synthesis technology is used in various situations, such as various information guidance systems such as a stock price guidance system by telephone, and reading out emails and Webs.

  As a conventional speech synthesis method, (1) a speech corpus including speech units of various units constituting a language such as phonemes, phonemes, words, etc., can be used for synthesis corresponding to the units in the input text. Search for a speech unit, select an optimal speech unit from the searched speech units, repeat this for all the units in the text, and connect the selected speech units as they are. There is a method of using synthesized speech (see Patent Document 1).

In this case, (2) there is a method in which each selected speech unit is connected to a synthesized speech after being subjected to signal processing for matching to the synthesis target prosody (see Non-Patent Document 1). ).
Japanese Patent No. 2761552 Satoshi TAKANO, Masanobu ABE "A NEW F0 MODIFICATION ALGORITHM BY MANIPULATING HARMONICS OF MAGNITUDE SPECTRUM", Eurospeech'99 Takano, Abe “Evaluation of the effect of prosody modification on sound quality by partial substitution experiment” Proceedings of the Acoustical Society of Japan 1-7-11, 2000 (3), pp. 217-218

  However, with the method (1) described above, it is possible to synthesize natural voice-like speech, but there is a problem that the prosody of the synthesized speech may be different from the synthesis target prosody, and (2) In the method, the prosody according to the synthesis target can be obtained, but there is a problem that the voice quality of the voice may be impaired by the signal processing.

  In other words, the conventional text-to-speech synthesis technique has a problem that the real voice quality of the synthesized speech and the accuracy of the prosody are in a trade-off relationship, and both cannot be satisfied at the same time.

  Thus, the current speech synthesis technology has not achieved sufficient quality compared to human speech, and there has been a strong demand for improving the quality of synthesized speech.

  SUMMARY OF THE INVENTION An object of the present invention is to realize a speech synthesis method and apparatus capable of providing a synthesized speech of higher quality than ever, according to the use situation of speech synthesis technology, particularly when the synthesized speech is reproduced. .

  As described above, there is a trade-off relationship between the real voice quality of synthesized speech and the accuracy of prosody, and there is a problem that sound quality deteriorates when attempting to synthesize the prosody accurately.

  Here, the degree of deterioration in sound quality when signal processing is performed in order to synthesize the prosody accurately differs depending on the prosody deformation amount and direction, the signal processing method, and the like (see Non-Patent Documents 1 and 2). .

  In addition, sound quality degradation of synthesized speech when signal processing is performed differs depending on whether it is heard with headphones or speakers, and naturally, it is better to listen only to synthesized speech with headphones. It is easy to understand the deterioration of the synthesized speech.

  In this way, the situation when playing back synthesized speech is whether it is played back through headphones, listened through speakers, or played back through speakers. There are various situations such as whether it is a noisy place such as a station premises, etc., and the way of understanding the deterioration of the quality of the synthesized speech differs depending on the situation and environment.

  In the present invention, in consideration of the situation when the synthesized speech is reproduced, in a situation where deterioration in sound quality due to signal processing is not conspicuous, signal processing is performed to correct the prosody, so that the sound quality due to signal processing can be improved for the user. By presenting the synthesized speech with the correct prosody without causing deterioration, the above-described problem of achieving both the sound quality of the synthesized speech and the correctness of the prosody is solved.

  ADVANTAGE OF THE INVENTION According to this invention, according to the utilization condition at the time of reproducing | regenerating synthetic | combination speech, it becomes possible to provide synthetic speech of higher quality than before.

  The headphone referred to in this application is used in direct contact with the body of one person, and includes all acoustic devices that transmit audio signals to the human ear, and the speaker referred to in this application is a human being. It includes all acoustic equipment that is used without touching the body and that transmits audio signals to one or more human ears, primarily through the air.

  The feature of the present invention is that the use of speech synthesis technology is taken into consideration, and whether or not prosodic deformation by signal processing is performed according to the situation when the synthesized speech is reproduced, so that the sound quality degradation is not noticeable. In a noisy environment, the prosody can be corrected and the voice can be played back, which is to synthesize higher-quality voice than before.

  As a result, it is possible to provide higher-quality synthesized speech for various information guidance services, etc., and speech synthesis technology can also be used in fields where synthesized speech could not be used due to quality problems. become.

  FIG. 1 shows an example of an embodiment of a speech synthesizer according to the present invention. In the figure, 1 is a database (DB), 2 is a database search unit (DB search unit), 3 is a speech unit selection unit, 4 Is a status determination unit, 5 is a signal processing unit, and 6 is a connection unit. Hereinafter, the operation of each unit will be described together with its configuration.

  DB1 is a speech corpus that includes speech units of various units constituting a language such as phonemes, phonemes, and words. Specifically, speech corpus, speech prosodic information, phoneme label sequence corresponding to utterance content, speech segment A speech corpus including information for synthesis such as label data indicating a boundary is stored.

  The DB search unit 2 is used for speech synthesis such as input text, phoneme series of the text obtained by a text analysis unit (not shown), prosody as a synthesis target, specification of a database to be used and a signal processing method. The input information 101 including the control information is used as an input, the speech unit usable for synthesis is searched from the database 1 for each unit in the text, and the speech unit selection unit 3 using the searched speech unit 102 as a search result. To pass.

  The speech unit selection unit 3 selects an optimal combination of speech units from the speech units 102 for each unit in the text passed from the DB search unit 2. The optimal speech element combination 103 is calculated by calculating, as a cost, elements related to the quality of synthesized speech such as prosody, phonological environment, connectivity, and linguistic information, and searching for a speech element combination that minimizes the cost. Is selected and passed to the situation determination unit 4 as a selection result.

  The situation determination unit 4 sends the optimum speech unit combination 103 passed from the speech unit selection unit 3 to the signal processing unit 5 or the connection unit 6 as described later according to the situation when the synthesized speech is reproduced. hand over.

  When the speech unit combination 103 is passed from the situation determination unit 4, the signal processing unit 5 corrects the prosody for each of the speech units included in the combination 103 to match the prosody to be synthesized. Signal processing is performed, and the result is passed to the connection unit 6 as a modified speech element combination 104.

  The connection unit 6 connects the speech unit combination 103 delivered from the situation determination unit 4 or the modified speech unit combination 104 delivered from the signal processing unit 5, and is reproduced as synthesized speech (data) 105 (not shown). Output to audio equipment for use.

  The configuration and operation of the DB1, the DB search unit 2, the speech unit selection unit 3, the signal processing unit 5 and the connection unit 6, and the text analysis unit (not shown) are the same as those of the existing speech synthesizer. Details are omitted.

  FIG. 2 shows the flow of processing in the situation determination unit 4, and the operation of the situation determination unit 4 will be described below in accordance with this.

  The situation determination unit 4 receives the optimum speech unit combination 103 from the speech unit selection unit 3, and also includes device identification information 106 and an acoustic device indicating whether the acoustic device for reproducing the synthesized speech is a headphone or a speaker. Is a speaker, receives the noise level (information) 107 around the installation location of the speaker, and determines from the device identification information 106 whether the acoustic device that reproduces the synthesized speech is a headphone or a speaker (s1). At this time, if it is a headphone, since the influence of the sound quality deterioration due to the signal processing is large, the optimum speech unit combination 103 is passed to the connection unit 6 (s2).

  On the other hand, when the device identification information 106 indicates that the acoustic device that reproduces the synthesized speech is a speaker, it is determined whether or not the noise level 107 is equal to or lower than a predetermined threshold value (s3). At this time, if it is less than the threshold value, it is assumed that the influence of the sound quality deterioration due to the signal processing is large, and the optimum speech unit combination 103 is passed to the connection unit 6 (s2). As a result, the optimum speech unit combination 103 is passed to the signal processing unit 4 (s4).

  Here, step s3 is omitted, and when the acoustic device that reproduces the synthesized speech is a speaker, it is assumed that the influence of the sound quality deterioration is always small and the optimum speech unit combination 103 is passed to the signal processing unit 4. Also good.

  Further, as the device identification information 106 described above, the specification information indicating whether the acoustic device input from the user is a headphone or a speaker, or the headphone jack in a device such as a personal computer or a television equipped with a headphone jack is connected to the headphone. A signal indicating whether or not the plug is connected can be used, and the noise level 107 is obtained when an output signal from a microphone or the like placed around the speaker installation place is passed through an appropriate integration circuit or the like. The resulting signal averaged to some extent on the time axis can be used.

The block diagram which shows an example of embodiment of the speech synthesizer of this invention Flow chart of processing in the situation determination unit in FIG.

Explanation of symbols

  1: Database (DB), 2: Database search unit (DB search unit), 3: Speech segment selection unit, 4: Situation determination unit, 5: Signal processing unit, 6: Connection unit.

Claims (2)

Using a database storing a speech corpus that includes speech units of units constituting a language, a computer searches the database for speech units that can be used for synthesis corresponding to the units in the input text. In the speech synthesis method, an optimal speech unit is selected from the retrieved speech units, this is repeated for all the units in the text, and the selected speech units are connected to form synthesized speech. ,
The calculator
Determine whether the audio device that plays the synthesized sound is a headphone or a speaker,
If the audio device is a headphone, the selected speech segments are connected as they are to produce synthesized speech,
If the audio equipment is a speaker , further determine whether the noise level around the speaker installation location is below a predetermined threshold,
If it is below the threshold, each selected speech segment is connected as it is to make a synthesized speech,
A speech synthesis method characterized in that if it is larger than the threshold value, the selected speech unit is connected to the selected speech segment after performing signal processing for correcting the prosody, and then synthesized speech. A database storing a speech corpus including speech units of units constituting a language, a database search unit for searching speech units usable for synthesis for each unit in the input text, and the search A speech unit selecting unit that selects an optimal speech unit from speech units for each unit in the text, and a speech unit corresponding to each unit in the selected text, A speech synthesizer provided with a connection unit for
A signal processing unit for performing signal processing for correcting the prosody for the speech unit;
Audio equipment to reproduce synthetic speech is determined whether headphones or speakers, if the headphone output as it is to the connecting portion of each speech unit selected by the speech unit selection unit, if the speaker of the speaker location It is determined whether or not the noise level in the surroundings is equal to or less than a predetermined threshold value. A speech synthesizer, comprising: a situation determination unit that outputs each speech unit selected by the unit to the connection unit via the signal processing unit.

Images