JPH04240900A - Speech converter - Google Patents

Abstract

PURPOSE:To convert an input voice into the voice of a character by a converter of simple constitution in a video system wherein a person who view a drama stands in for the characteristic of the drama. CONSTITUTION:The voice of a specific speaker is stored in a parameter A storage part 1 in advance. A voice analysis part 7 analyzes the voice parameter B of a voice inputted from an input terminal 5. A matching decision part 4 compares the parameters A and voice parameter B with each other. The time base is so adjusted according to the comparison result and a voice is synthesized by a voice synthesis part 3 by using the parameter A.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention particularly relates to a voice conversion device suitable for use in a video system capable of dubbing characters appearing in dramas.

0002

[Prior Art] There is a video system in which the lines of the characters in a drama are dubbed in place of the actors playing the characters, so that people watching the drama can freely dub the lines and enjoy the fun of participating in the drama. It is considered. That is, in such a video system, for example, in a scene where a predetermined character is speaking, the voice of the actor playing the character is muted, and the dialogue at that time is displayed as a telop. When a person watching the drama enters the lines while looking at this caption, the character's lines are output based on the input audio. In a video system having such a function, people watching a drama can enjoy the pleasure of actively participating in the drama.

[0003] In such a video system, it is desired that the voice of the input dialogue be converted into a sound quality that corresponds to the voice of the character. In other words, for example, when a man dubs a female character, the voice is converted into a female voice, and when a woman dubs a male character,
It is hoped that the voice will be converted into a male voice and dubbed. Furthermore, if the voice of the input dialogue is converted into a voice quality that is approximately the same as the character's voice, the user can enjoy the feeling of becoming a character in the drama. Therefore, it is conceivable to incorporate an audio converter into such a video system.

[0004] Conventionally, voice converters that convert a male voice into a female voice or a female voice into a male voice write the inputted voice to a memory and store the voice in this memory. A device is known in which audio is read out at a speed that corresponds to the pitch of the voice being converted.

That is, FIG. 6 is an example of a conventional audio converter. In FIG. 6, an audio signal is supplied to the input terminal 51. This audio signal is digitized by the A/D converter 52 and supplied to the memory 53. A write address AW is supplied to the memory 53 from an address generation circuit 54. This address generation circuit 54 is supplied with a write clock WCK. The input audio signal is stored in the memory 53 at the speed of this write clock WCK. Further, the memory 53 is supplied with a read address AR from an address generation circuit 55. The address generation circuit 55 is supplied with a read clock RCK. memory 53
The audio signal stored in the readout clock RCK
is read at a speed of The audio signal read from the memory 53 is sent to the output terminal 57 via the D/A converter 56.
is output from. In the audio converter as shown in FIG. 6, the read clock R is faster than the write clock WCK.
If the speed of CK is made faster, the input audio can be converted to high-pitched sounds, and if the speed of read clock RCK is slower than the speed of write clock WCK, the input audio can be converted to low-pitched sounds. can. Therefore, it is possible to make a male voice closer to a female voice, or to make a female voice closer to a male voice.

[0006]

[Problem to be solved by the invention] However, since the above-mentioned audio converter converts the height of all components on the frequency axis at a fixed ratio, it is difficult to convert the voice of the input dialogue into the voice of the character. It is impossible to convert to the same voice quality. Furthermore, the above-mentioned voice converter has a problem in that the converted voice becomes unnatural.

[0007] It is also conceivable to determine the characteristics of the input voice, synthesize the voice of the corresponding character from this voice, and output the synthesized voice. However, such processing is complex and difficult to implement. Further, even if an audio converter that performs such processing can be realized, such an audio converter will increase the circuit scale and cost.

[0008] Accordingly, an object of the present invention is to provide a voice converting device that can convert an input voice into the same voice as a character's voice with a simple configuration.

[0009]

[Means for Solving the Problems] This invention analyzes the voice information of a specific speaker in advance, records the voice information of the specific speaker and the voice parameters of the specific speaker, and analyzes the input voice information. Analyze the voice, compare the voice parameters of the input voice with the voice parameters of a specific speaker, and adjust the time axis at least to the time axis of the input voice information to adjust the voice according to the voice parameters of the specific speaker. This is a voice conversion device that outputs.

0010

[Operation] In the audio conversion used in such a video system, the conditions for audio conversion are specified as follows. (a). The lines are predetermined. (b). Since the voice is converted into the voice of a character, the characteristics of the voice after conversion are known in advance.

[0011] From this, the voice feature analysis parameter A of the voice of the character is stored in the parameter storage section 1 in advance, and the voice feature analysis parameter A stored in the parameter storage section 1 in advance is combined with the voice feature analysis parameter A of the input voice. By determining the matching with the voice feature parameter B and synthesizing the voice from the voice feature analysis parameters A stored in the parameter storage unit 1 so that the time axis matches the input voice, this voice can be You can obtain audio output that sounds like it has been converted into the voice of a character.

0012

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 shows an embodiment of the invention.

In FIG. 1, 1 is a parameter storage section. The parameter storage unit 1 stores in advance a time-series voice feature analysis parameter A (
Figure 3) is stored. This voice feature analysis parameter A is obtained by analyzing the frame period using, for example, linear predictive analysis (LPC). As shown in FIG. 3, marks M are attached to portions of the speech feature analysis parameters A that are easy to match, for example, frames that are stable for each voiced phoneme spanning multiple frames. Note that this parameter storage section 1 can be prepared, for example, in a predetermined area of a video disc.

That is, when creating a video disc in such a system, as shown in FIG.
1, the characters' lines are input in advance as audio signals. This audio signal is supplied to the audio analysis section 23 via the A/D converter 22. The voice analysis unit 23 performs LPC
The speech feature analysis parameter A is determined using the following. This audio feature analysis parameter A is recorded in advance in the parameter storage section 1 in a predetermined area of the video disc.

[0015] This speech feature analysis parameter A is
It may be changed as necessary to facilitate matching.

In FIG. 1, when performing voice conversion,
The speech feature analysis parameter A from the parameter storage section 1 is supplied to the speech synthesis section 3, and the matching judgment section 4
supplied to

Then, the voice of the dialogue is input to the input terminal 5 while being viewed on the playback screen. This audio signal is A/D
The signal is supplied to the speech analysis section 7 via the converter 6. The speech analysis section 7 analyzes the input speaker's speech feature analysis parameter B using the aforementioned LPC.

The speech feature analysis parameter A from the parameter storage section 1 and the speech feature analysis parameter B from the speech analysis section 7 are supplied to the matching determination section 4 . The matching determination unit 4 compares the information of one frame (see FIG. 3) or two frames from the beginning of the frame marked with the mark M among the audio feature analysis parameters A with the information of the audio feature analysis parameters B, The information of voice feature analysis parameter B is placed at the position of marker M of voice feature analysis parameter A with "
It determines whether the vehicle has passed, whether it has passed, whether it has passed, whether it has passed, or whether it has passed. Since the voice feature analysis parameter A stored in advance and the voice feature analysis parameter B analyzed from the input voice are the same lines, they can be interpreted as "not passed,""passing," or "passing." It is possible to easily make a judgment regarding "after".

The judgment output of the matching judgment section 4 is transmitted to the adjustment section 5.
supplied to The adjustment unit 9 adjusts the time axis of the output audio based on the time axis of the input audio. Further, the adjustment unit 9 changes the voice feature analysis parameter A in the parameter storage unit 1 based on the accent, intonation, etc. of the input voice.

The speech feature analysis parameter A from the parameter storage section 1 is sent to the speech synthesis section 3, and the speech synthesis section 3 synthesizes speech while compressing or expanding the time axis according to the output of the adjustment section 5. be done. This synthesized voice is D
/A converter 8 and is output from output terminal 9.

As described above, in one embodiment of the present invention, when a voice is input to the input terminal 5, the parameter storage section 1
The voice feature analysis parameter A stored in the parameter storage unit 1 is matched with the voice feature parameter B of the input voice. Speech is synthesized from the feature analysis parameters A. Therefore, when a voice is input to the input terminal 5, the output terminal 12 outputs a voice that appears to be a converted voice. Then, the voice feature analysis parameter A in the parameter storage section 1 is changed based on the accent, intonation, etc. of the input voice, so that the tone of the input voice can be retained in the converted voice.

The following methods can be considered for compressing and expanding the time axis. (a). As shown in FIG. 4A, the time axis is extended by repeating the same frame in the voice feature analysis parameter A (frame F2 is repeated in FIG. 4A), and as shown in FIG. 4B, the time axis is extended in the voice feature analysis parameter A. By thinning out the frames (frame F2 is thinned out in FIG. 4B), the time axis is compressed. (b). As shown in FIG. 5, the voice feature analysis parameter A
An interpolated frame (frame F in Figure 5) is inserted between the frames in
1 and frame F2)
By providing , the time axis is extended. (c). By varying the speed of the frame period in the audio feature analysis parameter A, the time axis is expanded or compressed.

[0023]

According to the present invention, the voice characteristic analysis parameter A of the voice of the character is stored in advance, and when the voice is input to the input terminal 5, the voice characteristic analysis parameter A stored in the parameter storage section 1 in advance is stored. Matching between parameter A and voice feature parameter B of the input voice is determined, and voice is synthesized from the voice feature analysis parameter A stored in the parameter storage unit 1 so that the time axis matches the input voice. be done. Therefore, when voice is input to input terminal 5,
The result is an audio output that sounds like this voice has been converted into the character's voice.

According to the present invention, complicated audio signal processing and calculations are not required, so the configuration does not become complicated and costs do not increase.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of an embodiment of the present invention.

FIG. 2 is a block diagram used to explain storage of voice feature analysis parameters A in an embodiment of the present invention.

FIG. 3 is a schematic diagram used to explain one embodiment of the present invention.

FIG. 4 is a schematic diagram used to explain one embodiment of the present invention.

FIG. 5 is a schematic diagram used to explain one embodiment of the present invention.

FIG. 6 is a block diagram of an example of a conventional audio converter.

[Explanation of symbols]

1 Parameter storage section 4 Matching judgment section 3 Speech synthesis section 5 Adjustment section 7 Speech analysis section

Claims (1)

[Claims] Claim 1: Speech-analyzing the voice information of a specific speaker, recording the voice information of the specific speaker and the voice parameters of the specific speaker, and voice-analyzing the input voice information;
The voice parameters of the input voice are compared with the voice parameters of the specific speaker, and at least the time axis is aligned with the time axis of the input voice information, and the voice is adjusted according to the voice parameters of the specific speaker. A voice conversion device that outputs.