JPH07210184A - Voice editor/synthesizer - Google Patents

Abstract

PURPOSE:To make it possible to synthesize a natural and smooth voice from unrecorded synthetic units when each of the units has a certain degree of recording quantity and to synthesize a natural voice generating no discontinuousness even if the synthetic unit is small. CONSTITUTION:A recording voice retrieving part 2, a connection unit determining part 3 and a sound signal connecting part 5 are prepared, and when a sentence of contents expressed a voice to be synthetized is inputted from a synthetic sentence signal input part 1, the retrieving part 2 retrieves the contents of recorded voices and selects several recording units including the voice concerned. The determining part 3 selects voices to be connected so as to quite connection distortion out of those voices, the connecting part 5 connects the selected voices and a synthetic signal output part 6 synthesizes a required voice.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a recording / editing / synthesizing device for synthesizing digitally recorded voices by editing, which is used for station guide broadcasting and operation explanation of electric appliances.

[0002]

2. Description of the Related Art Conventionally, this type of voice editing / synthesizing apparatus records a voice uttered by a person in advance in units of words, phrases, sentences, etc., and reads and edits the voice as necessary.
Synthesized into speech such as sentences and output.

That is, as in the example of the sentence 1 below, in order to output a voice saying "I will soon go to Line 3 for Tokyo", I will say "Soon → A3 → B2 → C1 → I will come" A desired sentence can be synthesized by selecting and outputting in sequence.

[0004] sentence 1: Mairi is soon to ○ Line ○ ○○○ go. A: Line B: Train type C: Destination A1: Line 1 B1: Stop at each station C1: Tokyo bound A2: Line 2 B2: Express C2: Yokohama bound A3: Line 3 B3: Rapid C3: Shinagawa bound B4: Rapid commuting C4: To Kawasaki

As described above, even in the above-mentioned conventional method, a desired voice can be synthesized by combining and outputting the synthesis units.

[0006]

However, in the above-described conventional voice editing / synthesizing apparatus, the synthesis unit is a sentence or a phrase, and since the voice is output for each unit, the recording amount is large and the sentence is fixed to a certain degree. Unless it can be synthesized. In addition, there is a problem that a voice that is not recorded as a synthesis unit cannot be output. That is, if the pattern of the sentence is changed or the destination of the train is changed as in the example of the following sentence, even if there are voices such as "To Tokyo" or "To Yokohama" in sentence 1, "To Tokyo" No '
Since there is no voice for "Yokohama", we have to record them newly.

Sentence 2: Soon, a train bound for ○○ will arrive on the ○○ line . A: Line D: Direction E: Destination A1: Line 1 B1: Odawara C1: Tokyo A2: Line 2 B2: Yokohama C2: Shizuoka A3: Line 3 B3: Shinagawa C3: Nagoya B4: To Kawasaki C4: To Aomori

In order to avoid such a new recording, the composition units are "Tokyo", "Yokohama", "Gono" and "Goga".
For example, if you reduce the number of words, connectives, particles, syllables, etc. instead of sentences or phrases, you can increase the vocabulary and sentences that can be synthesized without increasing the recording amount.
When playing back, only the synthetic units of "Yokohama" and "direction" are output in order, resulting in discontinuous voice such as "Yokohama, direction", which deteriorates the voice quality. There was a problem.

The present invention solves the above-mentioned conventional problems, and even if a synthesis unit that is not recorded has a certain amount of voice, it synthesizes a natural and smooth voice from those voices, or synthesizes them. An object of the present invention is to provide a voice editing / synthesizing device capable of synthesizing a natural voice in which discontinuity does not occur even if the unit is reduced.

[0010]

SUMMARY OF THE INVENTION In order to achieve the above object, the present invention uses a silent segment of an unvoiced plosive sound, a frictional segment of an unvoiced fricative sound, and the same voiced sound in the middle of continuous voice connection. It is provided with a means for connecting in sections where the spectra are similar.

According to the present invention, when a voice is connected, the connected voice signal linearly attenuates a certain section power from the connection point, and the connected voice signal linearly increases the same section power. Then, the signals of that section are added together.

In the present invention, when connecting voices, when the connecting place is a place where the pitch of the voices exists, in order to avoid the connection distortion of the voices due to the phase difference of the pitches, two voices of the connecting place are connected. The signal is correlated, and a portion having a high correlation is set as a connection point to reduce connection distortion due to the phase of voice.

[0013]

The present invention has the following actions due to the above-mentioned structure. That is, the amount of recording can be increased by connecting at a place where no distortion occurs due to the connection, such as a silent section of an unvoiced plosive while the voice is continuing, a friction section of an unvoiced fricative, or a section where the spectrum is similar for the same voiced sound. It is possible to synthesize a natural voice without discontinuity without increasing the.

In order to improve the voice connection,
The connected voice is weighted linearly from 1 to 0 in a certain section, and the connected voice is linearly changed from 0 to 1 in a certain section from the connection point. By weighting toward and adding two voices in the section, the voice connection at the connection point is improved, and a smooth voice can be reproduced.

Further, when connecting at a place where a pitch exists, by correlating two voices in the vicinity of the connecting place and setting a place having the largest correlation as a connecting place,
It is possible to determine a connection point that can reproduce natural sound with little phase connection distortion.

[0016]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of an embodiment of a voice editing / synthesizing apparatus of the present invention. In FIG. 1, 1 is a synthetic text signal input unit, 2 is a recorded voice search unit, 3
Is a connection unit determining unit, 4 is a voice signal reading unit, 5 is a voice signal connecting unit, 6 is a synthesized signal output unit, and 7 is recorded voice.

Speech generally has some features depending on the type of phoneme. That is, t∫ of unvoiced plosives or unvoiced affrications such as p, t, k has a silent closing sound at the beginning of the phoneme, and unvoiced fricatives s, f, h and ts of asexual affricate.
(Tsu) begins with a persistent, sharp, frictional sound. In the vowel section of speech, the spectrum change is constant to some extent.
Since the nature of the voice is constant in such a part, even if the voice is connected in the middle of the voice, the distortion due to the connection is not significant. Therefore, when a sentence having a voice content to be synthesized is input in the synthetic sentence signal input unit 1 of FIG. 1, the recorded voice search unit 2 searches the content of the recorded voice, and a synthesis unit including the voice. Select some. Then, the connection unit determination unit 3 selects a voice that can be connected with the least connection distortion from those voices, and outputs a signal corresponding to the voice selected by the voice signal reading unit 4 in the recorded voice 7. The voice signal is connected to the voice signal connection unit 5, and the synthesized signal output unit 6 outputs the connected voice signal.

According to the examples shown in FIGS. 2 and 3, when synthesizing the phrase "Shimoda area", first, the recording voice search unit 2 selects "Shimoda bound" including "Shimoda" of "Shimoda area". In addition, “Tokyo direction” including “Shimoda direction”, “Koufu direction”, “Hiratsuka direction”, and “Atami direction” are selected. Next, the connection unit determination unit 3 determines “Shimoda bound” and “Hiratsuka direction”, in which a desired voice is synthesized by connecting with the vowel a, from the selected voices, and the voice signal connection unit 5 Then, the vowels a of these voices are connected to synthesize "Shimoda area", and the synthesized voice is output by the synthetic signal output unit 6.

Further, in the case of connecting voices by the voice signal connection unit 5, as shown in FIG. 4, in order to improve the connection between the two voices, a certain interval such as a section corresponding to the pitch cycle at the connecting point is provided. The voice A on the side connected to the voice data of the section is weighted from 1 to 0, the voice B on the connecting side is weighted from 0 to 1, and the voices of the weighted sections are added to each other. By doing so, it is possible to improve the continuity of voice and reduce distortion due to connection.

Further, when the connection is made at a place where a pitch exists, distortion may occur due to the phase shift of the pitch of the voice. Therefore, when the connected portion is determined, as shown in FIG. 5, the correlation between the two voices is obtained from the following equation, and the connection point at the location having the highest correlation is moved to obtain the phase shift (shift expression). It is possible to match the phase of the voice by correcting the amount) and avoid the connection distortion due to the difference in the phase of the voice.

[0021]

[Equation 1]

As described above, according to this embodiment, even if the voice is not recorded, the voice is naturally recorded from the recorded voice to some extent by connecting the voice at a place where the voice can be easily connected even in the middle of the voice. You can synthesize smooth voices with.

Further, when recording, a voice including a preceding vowel, a voice delimited by a silent part of an unvoiced plosive, or a voice delimited by a frictional part of an unvoiced fricative is recorded as a unit of a voice starting at a part which is easily connected. In addition, by connecting these parts at the time of reproduction, it is possible to synthesize a voice having a quality equivalent to that of editing / synthesis in sentences or clauses with a recording amount equivalent to that of editing / synthesis in units of words, connecting words, and particles.

[0024]

As is apparent from the above embodiment, the present invention provides a silent section of a voiceless plosive in the middle of continuous voice.
By connecting at a place where distortion due to the connection does not occur, such as a friction section of unvoiced fricative, a place where the spectrum is similar for the same voiced sound, natural speech that does not cause discontinuity without increasing the recording amount is synthesized. be able to.

In order to improve the connection of voice,
The connected voice is weighted linearly from 1 to 0 in a certain section, and the connected voice is linearly changed from 0 to 1 in a certain section from the connection point. By weighting toward and adding two voices in the section, the voice connection at the connection point is improved, and a smooth voice can be reproduced.

Further, when connecting at a place where a pitch exists, by correlating two voices in the vicinity of the connecting place and setting a place having the largest correlation as a connecting place,
It is possible to determine a connection point that can reproduce natural sound with little phase connection distortion.

[Brief description of drawings]

FIG. 1 is a schematic block diagram showing the configuration of a voice editing device according to an embodiment of the present invention.

FIG. 2 is a schematic diagram for explaining the operation of the device.

FIG. 3 is a waveform diagram for explaining the operation of the device.

FIG. 4 is a waveform diagram for explaining the operation of the device.

FIG. 5 is a waveform diagram for explaining the operation of the device.

[Explanation of symbols]

 1 Synthetic text signal input unit 2 Recorded voice search unit 3 Connection unit determination unit 4 Voice signal reading unit 5 Voice signal connection unit 6 Synthetic signal output unit 7 Recorded voice

Claims (3)

[Claims] 1. A voice editing / synthesizing device for outputting voices recorded in units of sentences or phrases in combination, in which voices are connected by a silent segment of a silent plosive and a frictional segment of an unvoiced fricative in the middle of continuous voice. And a voice editing / synthesizing device having means for connecting in sections having similar spectra. 2. When connecting voice, the connected voice signal linearly attenuates power in a certain section from the connection point, and the connected voice signal linearly increases power in the same section, 2. The voice editing / synthesizing apparatus according to claim 1, wherein signals of the section are added together. 3. When connecting voices, when the connecting place is a place where the pitch of the voice exists, in order to avoid the connection distortion of the voice due to the phase difference of the pitch, the correlation of two voice signals at the connecting place. The voice editing / synthesizing apparatus according to claim 1, wherein a portion having a high correlation is set as a connection point to reduce connection distortion due to a voice phase.