JPH11249679A - Voice synthesizer - Google Patents

Abstract

PROBLEM TO BE SOLVED: To bring a synthetic voice when a routine phrase is read out from an inputted character line with the synthetic voice near to a natural voice. SOLUTION: In a voice synthesizer provided with a phoneme continuous time storage part M2, a basic frequency storage part M3, an amplitude storage part M4 respectively storing a phoneme continuous time length series, a basic frequency series and an amplitude or power series extracted from the phrase uttered by a human being, and provided with a phoneme piece storage part M1 storing a phoneme or a phoneme chain as a phoneme unit, and phoneme information as the phoneme piece, a phoneme piece series read out from a phoneme piece storage part 2 is expanded/contracted according to the phoneme continuous time length series read out from a rhythm pattern storage part 1 to be connected, and a basic frequency is added according to the basic frequency series read out from the rhythm pattern storage part 1, and at this time, the normalization is performed so as to enter in the basic frequency range, and further, the amplitude is added according to the amplitude or power series read out from a rhythm pattern storage part, and the voice is synthesized.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesizing apparatus which can be used for converting text in which a fixed text and a non-fixed text are mixed into speech.

[0002]

2. Description of the Related Art Japanese Unexamined Patent Publication No. Hei 5-27789 discloses a "voice synthesizer". This is to perform overlap processing at a connection portion when mixed voices of different systems such as rule synthesis, analysis synthesis, and recording / reproduction are used together. Japanese Patent Application Laid-Open No. 8-63187 discloses a "speech synthesis apparatus" that uses a fundamental frequency and a phoneme time length extracted from natural speech for a fixed part.

[0003]

Conventionally, there are a recording and editing system and a rule synthesizing system as a synthesizing system of a voice synthesizing device. In the former, an announcer or the like registers voice for each phrase and selects and combines them as appropriate to create a message.There is a possibility that a good voice close to the real voice may be obtained, but the data volume is large. There are problems such as the inability to cope with outside phrases and the need to secure the same speaker to add a new phrase. On the other hand, the latter accumulates speech data in fine units such as phonemes and syllables and enables synthesis of arbitrary vocabulary.However, the sound quality is inferior to the recording and editing method, especially the fundamental frequency, phoneme duration, amplitude, etc. will inevitably those mechanical and unnatural for the prosody pattern imparted by the rules, such as.

For this reason, a standard message that does not require output of an arbitrary vocabulary uses a recording / editing method with good sound quality, and a rule synthesis is used in a case where voice conversion from text is required. However, there are many applications in which an arbitrary vocabulary is partially embedded in a fixed sentence, such as a place name being embedded in a fixed message by voice guidance of car navigation. In such a case, for all of the optional vocabulary, reduce the overall sound quality and adopt rule synthesis, or give up the output of the optional vocabulary and use the recording and editing method,
Alternatively, a choice must be made as to whether to adopt a mixed system in which the standard part is recorded and edited and only the arbitrary vocabulary part is subjected to rule synthesis.

[0005] A problem when recording editing and rule synthesis are mixed is that the two systems have completely different output voice qualities, so that not only is it uncomfortable to hear but also very difficult to hear. JP-A-5-27789
The "voice synthesizer" described in Japanese Patent Application Laid-Open No. HEI 10-125555 addresses this problem by overlapping output voices between different systems and connecting them. However, even in this case, it is inevitable that the speaker changes between the fixed part and the arbitrary word part, and the basic problem is not solved. Further, in the overlap portion, two speakers may be talking at the same time, so that it may be difficult to hear.

On the other hand, in the "speech synthesis apparatus" described in Japanese Patent Laid-Open No. 8-63187, phoneme parameters are generated by connecting phonemes or syllables and the like to a fixed sentence in a regular synthesis. By adding the fundamental frequency and phoneme duration extracted from the voice, naturalness is improved while maintaining speaker continuity with an arbitrary word portion. However, the various prosodic and phonological parameters are related to each other, and only a part (fundamental frequency and phonological duration) of a set of rules that are constructed to balance as a whole has completely different speaker characteristics and utterances. Porting from stylized speech can create unexpected inconsistencies and undermine overall naturalness. For example, the fundamental frequency may drop significantly toward the end of the sentence. At this time, if the amplitude is not sufficiently reduced, an unnaturally low voice will stand out. In such a case, the opening of the mouth is originally small, and the voice spectrum itself should be changed. Using too clear phoneme segment data also gives a sense of incongruity. In addition, there is a range of fundamental frequencies that can be handled in phoneme segment data, and the fundamental frequency pattern of natural speech usually has a wider dynamic range than this. There is fear. Therefore, an object of the present invention is to solve such a problem and improve the naturalness of a synthesized speech of a fixed phrase.

[0007]

According to the first aspect of the present invention, there is provided a phoneme duration sequence extracted from a phrase uttered by a human,
A phoneme duration storage unit, a fundamental frequency storage unit, an amplitude storage unit, and a phoneme fragment storage that stores a phoneme or a phoneme chain as a phoneme unit and stores phoneme information as a phoneme segment. The phoneme sequence read and arranged from the phoneme storage according to the input character string is expanded and contracted according to the phoneme duration sequence read from the prosody pattern storage, and read from the prosody pattern storage. This is a speech synthesizer that performs basic frequency assignment according to a basic frequency sequence, and performs amplitude assignment according to the amplitude or power sequence read from the prosody pattern storage unit to synthesize speech.

According to a second aspect of the present invention, in the voice synthesizing apparatus according to the first aspect, a basic frequency range storage unit that stores a range of a basic frequency that can be reasonably synthesized by the phoneme unit set stored in the phoneme unit storage unit. The voice synthesizer includes a basic frequency sequence read from the basic frequency storage unit, and performs normalization so as to fall within the basic frequency range read from the basic frequency range storage unit to synthesize a voice.

According to a third aspect of the present invention, in the speech synthesizer according to the first aspect, the phoneme unit storage unit stores a plurality of phoneme units for each fundamental frequency range to be adapted to one phoneme unit. A speech synthesis apparatus for selectively using speech segments corresponding to the fundamental frequency read from the fundamental frequency storage unit to perform speech synthesis.

According to a fourth aspect of the present invention, in the voice synthesizing apparatus according to the first aspect, the phoneme unit storage unit stores a plurality of phoneme units for each amplitude range to be adapted to one phoneme unit. This is a speech synthesizer that performs speech synthesis by selectively using phonemes corresponding to the amplitude or power read from the fundamental frequency storage unit.

According to a fifth aspect of the present invention, in the voice synthesizing apparatus according to the first aspect, the phoneme segment storage unit stores a plurality of phoneme segments for each phoneme duration range to be adapted to one phoneme unit. This is a speech synthesizer that performs speech synthesis by selectively using a phoneme segment corresponding to the phoneme duration stored in the phoneme duration stored in the phoneme duration storage unit.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the speech synthesizer according to the present invention will be described. FIG. 1 shows a configuration in this embodiment. In FIG. 1, M1 is a phoneme segment storage unit that stores phonemes or phoneme chains as phoneme units and phoneme information as phoneme segments, and stores a plurality of phoneme segments for each basic frequency range to be applied to one phoneme unit. The phoneme segment corresponding to the fundamental frequency stored and read from the fundamental frequency range storage unit M5 is selectively used. M2, M3, and M4 are respectively a phoneme duration time storage unit, a fundamental frequency storage unit, a fundamental frequency storage unit, and a phoneme duration time length sequence, a fundamental frequency sequence, and an amplitude or power sequence extracted from a phrase uttered by a human. 4 shows a storage unit. M5 is a fundamental frequency range storage unit, which stores a range of fundamental frequencies that can be reasonably synthesized by the phoneme unit set stored in the phoneme unit storage unit M1.

The operation of each section will be described below. The prosody pattern selection unit 1 selects each of the phoneme duration time, the fundamental frequency, and the amplitude from the input prosody pattern ID (identifier for identifying the prosody pattern: for example, a pattern corresponding to the number by a number or the like). . The phoneme unit selection unit 2 obtains a phoneme unit label from the input character string, and refers to the range of each prosody pattern of the phoneme duration, the fundamental frequency, and the amplitude selected by the prosody pattern selection unit 1 to refer to these information. A necessary phoneme segment is retrieved from the phoneme segment storage unit M1.

FIG. 2 shows an example of the data structure of the phoneme segment storage unit M1. Same phoneme label, for example,
A plurality of data of different applicable prosodic parameter ranges are stored for “A”. As shown here, the prosody parameter range may be categorized, for example, the length of the time length range, the size of the amplification range, or may indicate the lower limit and upper limit like the fundamental frequency range, or may be a mixture of them. You can do it. In the data column in the table, phoneme piece data, waveform data, and spectrum parameters are actually stored. From these, the one with the same prosody parameter range that has the same label is selected.

The phoneme unit expansion / contraction connection unit 3 converts the phoneme unit sequence selected by the phoneme unit selection unit 2 according to the input character string into
Referring to the range of the phoneme duration selected by the prosody pattern selection unit 1, the phoneme segments are connected by expanding and contracting according to the phoneme duration pattern selected from the phoneme duration storage unit M2. In the fundamental frequency range storage unit M5,
The basic frequency range covered by the entire phoneme unit data set stored in the phoneme unit storage unit M1 is stored, and the basic frequency pattern normalizing unit 4 stores the basic frequency pattern selected from the basic frequency range storage unit M5. If it is outside of this range, the selected fundamental frequency pattern is normalized according to this range. The fundamental frequency assigning unit 5 assigns a normalized fundamental frequency pattern to the phoneme unit sequence pattern connected by the phoneme unit expansion / contraction connection unit 3.

The amplitude assigning unit 6 refers to the connected phoneme unit sequence pattern to which the fundamental frequency has been assigned by referring to the range of the amplitude pattern selected by the prosody pattern selecting unit 1.
A synthesized voice is created by adding the amplitude pattern selected from the amplitude storage unit M4. Note that a general technique of regular speech synthesis can be used for the expansion and contraction and connection of the phonemic segments, and the provision of the fundamental frequency and the amplitude, and thus detailed description is omitted here.

[0017]

According to the first aspect of the present invention, three basic prosody parameters, namely, a fundamental frequency, a phoneme duration, and an amplitude, are extracted from the same friendly speech phrase, thereby obtaining a prosodic parameter between the prosodic parameters. The mismatch can be suppressed, and the naturalness of the synthesized speech can be improved.

According to the second aspect, by normalizing the fundamental frequency to a range that can be supported by the phoneme unit database, it is possible to prevent the fundamental frequency from being applied unreasonably and to prevent the clarity of the synthesized speech from lowering.

According to the third aspect of the present invention, by selectively using the speech segment data corresponding to the fundamental frequency to be imparted, a fundamental frequency of a natural voice having a wide dynamic range can be imparted, and clarity can be improved. Naturalness can be improved without dropping.

According to the fourth aspect of the present invention, if only the amplitude is changed using the same phoneme data, a mechanical change occurs as if the volume of a speaker is manipulated, and the naturalness is impaired. By selectively using phoneme segment data, it is possible to change the phonological characteristics according to the size of the voice, and it is possible to apply a wide amplitude change in the dynamic range of natural voice, and obtain a synthesized voice closer to human voice. Can be

According to the fifth aspect of the present invention, by selectively using phoneme piece data corresponding to the phoneme duration to be set, a characteristic part of a consonant is lost due to excessive truncation of a phoneme piece, or a short vowel is generated. It is possible to avoid the mechanical sound quality due to the repetition of the stationary part, and it is possible to give a tempo change with a wide dynamic range of the natural sound without deteriorating the clarity, and to obtain a synthesized voice closer to human voice.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a speech synthesis device according to the present invention.

FIG. 2 shows a data structure of a phoneme segment storage unit shown in FIG.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Prosody pattern selection part, 2 ... phoneme piece selection part, 3 ... phoneme piece expansion / contraction connection part, 4 ... Fundamental frequency pattern normalization part, 5 ...
Basic frequency assigning unit, 6: amplitude assigning unit, M1: phoneme unit storing unit, M2: phoneme duration storing unit, M3: fundamental frequency storing unit, M4: amplitude storing unit, M5: basic frequency range storing unit.

Claims (5)

[Claims] 1. A phoneme duration storage unit, a fundamental frequency storage unit, an amplitude storage unit that stores a phoneme duration sequence, a fundamental frequency sequence, and an amplitude or power sequence extracted from a phrase uttered by a human. A phoneme unit storage unit that stores phonemes or phoneme chains in phoneme units and stores phoneme information as phoneme segments. A phoneme unit sequence read from the phoneme unit storage unit according to the input character string, and a phoneme continuation read from the prosodic pattern storage unit. Connect by expanding and contracting according to the time length sequence, perform the fundamental frequency assignment according to the fundamental frequency sequence read from the prosody pattern storage unit, and synthesize the voice by applying the amplitude according to the amplitude or power sequence read from the prosody pattern storage unit A speech synthesizer characterized in that: 2. The speech synthesizer according to claim 1, further comprising: a fundamental frequency range storage unit that stores a range of fundamental frequencies that can be reasonably synthesized by the phoneme unit set stored in the phoneme unit storage unit. A speech synthesizer characterized by performing normalization on a fundamental frequency sequence read from a storage unit so as to fall within the fundamental frequency range read from the fundamental frequency range storage unit to synthesize speech. 3. The speech synthesizer according to claim 1, wherein said phoneme segment storage unit stores a plurality of phoneme segments for each fundamental frequency range to be adapted to one phoneme unit. A speech synthesis apparatus characterized in that speech synthesis is performed by selectively using phonemes corresponding to a fundamental frequency read from a frequency storage unit. 4. The speech synthesizer according to claim 1, wherein said phoneme unit storage unit stores a plurality of phoneme units for each amplitude range to be adapted to one phoneme unit. A speech synthesis apparatus characterized in that speech synthesis is performed by selectively using phonemes corresponding to amplitude or power read from a unit. 5. The speech synthesizer according to claim 1, wherein said phoneme segment storage unit stores a plurality of phoneme segments for each phoneme duration length range to be adapted to one phoneme unit. A speech synthesizer characterized in that speech synthesis is performed by selectively using phoneme segments corresponding to the phoneme duration read out from the phoneme duration storage.