KR101706123B1 - User-customizable voice revision method of converting voice by parameter modification and voice revision device implementing the same - Google Patents

Abstract

The present invention relates to a user-customized voice correction method for converting a tone color by parameter change and a voice correction apparatus for implementing the same. More specifically, the user-customized voice correction apparatus includes: (1) receiving raw voice data; (2) extracting speech parameters including pitch, characteristic waveform (CW), power and line spectrum frequency (LSF) from the input original speech data; (3) changing, with respect to the extracted speech parameters, first characteristics of a plurality of predetermined specific sounds to second features having a predetermined normal speech; And (4) synthesizing the corrected voice data based on the voice parameters having the changed second characteristic.
According to the user-customized speech correction method and the speech correction apparatus embodying the present invention, the user-customized speech correction apparatus receives the original speech data and extracts the pitch , The characteristic waveform (CW), the power and the line spectrum frequency (LSF), extracts first characteristics of a predetermined plurality of specific sounds with respect to the extracted speech parameters, Speech, breathy speech, and nasal speech by synthesizing the corrected speech data based on the speech parameters having the modified second characteristic, It is possible to analyze and improve specific voices such as voice, and change the tone to a cold feeling or a warm feeling.
In addition, it is possible to change a voice which is difficult to be identified through voice communication or an unsatisfactory voice of a speaker to a voice of a clear or desired tone color, assists a physical problem of a vocal institution of the speaker, It is possible to variously correct the voice according to the user's desire.

Description

TECHNICAL FIELD [0001] The present invention relates to a user-customized voice correction method for converting a tone color by parameter change, and a voice correction device for implementing the same. [0002]

More particularly, the present invention relates to a user-customized voice correction method for converting a tone color by parameter change and a voice correction apparatus for implementing the same.

The voice codec is a combination of a coder / encoder for converting a voice signal into a digital signal and a decoder for converting the voice signal to a coder / encoder in a reverse manner. The voice codec is a combination of MP3, AC3, AAC, OGG and WMA Codec technology is known.

A method of waveform interpolation (WI), which is one kind of speech codec, is a process of extracting parameters of speech and synthesizing speech as shown in FIG. 1 is a diagram conceptually illustrating a process of executing a waveform interpolation method among voice codecs according to the related art. As shown in FIG. 1, the waveform interpolation method according to the related art performs a process of reducing the amount of data by quantizing the original parameters since the bandwidth that can be transmitted for voice communication is limited. More specifically, when the digitized voice is input to the analyzer (analyzer-synthesizer layer) on the encoder side, the parameters of the input voice are quantized (quantization layer). After quantization, the parameters are dequantized again (quantization layer), passed to the synthesizer, and the digitized voice is output from the synthesizer.

In this regard, Korean Patent Registration No. 10-0768090 (Oct. 7, 2007) discloses a waveform interpolation method and apparatus for reducing the amount of calculation of reordering parameters in decoding, and Korean Patent Application Publication No. 10-2001- 0087391 (Sep. 15, 2001) discloses a time synchronous waveform interpolation method for a speech segment that minimizes the bits required for encoding.

However, in these conventional techniques, studies are mainly conducted to quantify the speech quality by using the values of the irregular component such as jitter, shimmer, harmonic-to-noise ratio (HNR) However, a technique for correcting the user's voice appropriately is not disclosed. Further, a technique for correcting the user's general voice at various angles has a problem not disclosed.

The present invention has been proposed in order to solve the above-mentioned problems of the previously proposed methods, and a user-customized speech correction apparatus receives raw speech data and extracts pitch, characteristic waveform (CW), power And a line spectrum frequency (LSF), and for the extracted speech parameters, changing the first characteristics of a plurality of predetermined specific sounds to the second characteristics having a predetermined normal speech, By synthesizing the corrected speech data based on the speech parameters having the second characteristic, a specific voice such as hoarse speech, rough speech, breathy speech and nasal speech can be analyzed and / User-customizable voices that can change the tone by changing the parameter, which can be improved and change the voices to a cold feeling or warm feeling The method and to provide an audio correction device to implement this as its object.

Further, the present invention can change the voice which is difficult to be identified through voice communication or the unsatisfactory voice of the speaker to a voice of a clear or desired tone color, assists a physical problem of a vocal organ of a speaker, There is provided a user-customized voice correction method for converting a tone color by parameter change and capable of correcting various voices in accordance with a user's desire by causing a user to generate a voice of a tone to be appealed, and a voice correction device implementing the same For other purposes.

According to an aspect of the present invention, there is provided a method of converting a tone color into a user-

A user-

(1) receiving raw voice data;

(2) extracting speech parameters including pitch, characteristic waveform (CW), power and line spectrum frequency (LSF) from the input original speech data;

(3) changing, with respect to the extracted speech parameters, first characteristics of a plurality of predetermined specific sounds to second features having a predetermined normal speech; And

(4) synthesizing the corrected voice data based on the voice parameters having the changed second characteristic.

Preferably, in said step (3)

The specific voice may comprise at least one selected from the group consisting of a hoarse voice, a rough voice, a breathy voice, and a nasal voice.

More preferably, in the step (3)

A change to the second features of the first features,

And stabilizing the pitch of the specific voice within a preset range.

More preferably, in the step (3)

A change to the second features of the first features,

And reconstructing a spectral envelope by correcting a line spectrum pair (LSP) of the specific speech.

More preferably, in the step (3)

A change to the second features of the first features,

Separating the characteristic waveform of the specific speech into a smoothly evolving waveform (SEW) component and a rapidly evolving waveform (REW) component, and adjusting the separated SEW component and REW component within a predetermined range .

Preferably,

Between steps (1) to (4) above,

And the step of quantizing the raw speech data is not included.

Preferably, in the step (3)

(3-1) changing the extracted voice parameters from a preset first tone color range to a predetermined second tone color range,

One of the first tone color range and the second tone color range may be a tone color range of a cold feeling and the other tone color range may be a warm tone color range.

According to the user-customized speech correction method and the speech correction apparatus embodying the present invention, the user-customized speech correction apparatus receives the original speech data and extracts the pitch , The characteristic waveform (CW), the power and the line spectrum frequency (LSF), extracts first characteristics of a predetermined plurality of specific sounds with respect to the extracted speech parameters, Speech, breathy speech, and nasal speech by synthesizing the corrected speech data based on the speech parameters having the modified second characteristic, It is possible to analyze and improve specific voices such as voice, and change the tone to a cold feeling or a warm feeling.

In addition, it is possible to change a voice which is difficult to be identified through voice communication or an unsatisfactory voice of a speaker to a voice of a clear or desired tone color, assists a physical problem of a vocal institution of the speaker, It is possible to variously correct the voice according to the user's desire.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 conceptually illustrates the execution of a waveform interpolation method among speech codecs according to the prior art; FIG.
2 illustrates a flow of a user-customized speech correction method for converting a tone color by parameter change according to an embodiment of the present invention.
3 is a conceptual view illustrating a user-customized voice correction method for converting a tone color by parameter change according to an embodiment of the present invention.
FIG. 4 is a block diagram illustrating a process performed in a WI analyzer of a user-customized voice correction method for converting a tone color by parameter change according to an exemplary embodiment of the present invention; FIG.
FIG. 5 is a block diagram illustrating a process performed in a WI synthesizer of a user-customized speech correction method for converting a tone color according to an exemplary embodiment of the present invention; FIG.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily carry out the present invention. In the following detailed description of the preferred embodiments of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear. The same or similar reference numerals are used throughout the drawings for portions having similar functions and functions.

In addition, in the entire specification, when a part is referred to as being 'connected' to another part, it may be referred to as 'indirectly connected' not only with 'directly connected' . Also, to "include" an element means that it may include other elements, rather than excluding other elements, unless specifically stated otherwise.

The present invention relates to a user-customized voice correction method for converting a tone color by parameter change and a voice correction device for implementing the same, and a user-customized voice correction device for converting a voice tone by changing a parameter according to the feature of the present invention, A memory to be stored, and a microprocessor for processing the stored voice data. For example, such a voice correction device may include a portable terminal electrically connected to a microphone, a communication terminal, a personal computer, a notebook, a PDA, a smart phone, a tablet PC, an MP3 player, and the like. The processing of the voice data performed by the voice correction apparatus will be described in detail through a user-customized voice correction method described later.

FIG. 2 is a flowchart illustrating a user-customized voice correction method for converting a tone color by parameter change according to an exemplary embodiment of the present invention. Referring to FIG. As shown in FIG. 2, a user-customized voice correction method for converting a tone color by parameter change according to an embodiment of the present invention includes receiving (S510) raw voice data, (S530) extracting voice parameters including a characteristic waveform (CW), power and a line spectrum frequency (LSF), extracting first characteristics of a predetermined plurality of specific sounds with respect to the extracted voice parameters, (S550), and synthesizing the corrected voice data based on the voice parameters having the changed second characteristic (S570). 2, in the user-customized speech correction method for converting a tone color by parameter change according to an embodiment of the present invention, unlike FIG. 1, a step of quantizing raw speech data and dequantizing it again not included. Hereinafter, each step of the user-customized voice correction method for converting a tone color by parameter change according to an embodiment of the present invention will be described in more detail with reference to the accompanying drawings.

3 is a conceptual diagram illustrating a user-customized voice correction method for converting a tone color by changing a parameter according to an embodiment of the present invention. As shown in FIGS. 2 and 3, in step S510, the user-customized voice correction apparatus can receive the raw voice data. The inputted raw voice data may include a characteristic that can be distinguished as a specific voice. For example, raw speech data may include features such as hoarseness, breath, harshness, nasalance, and the like. Then, in step S530, the user-customized speech correction apparatus can extract speech parameters including pitch, characteristic waveform (CW), power and line spectrum frequency (LSF) from the inputted original speech data. The process of extracting the voice parameters from the raw voice data, as in steps S510 and S530, will be described in more detail with reference to FIG.

FIG. 4 is a block diagram illustrating a process performed by a WI analyzer in a user-customized speech correction method for converting a tone color by parameter change according to an exemplary embodiment of the present invention. 4, the WI analyzer 100 analyzes (130) LPC (linear predictive coding), interpolates a line spectral frequency (LSF) (120) for the inputted raw speech data, May be provided to the LP (linear prediction) analysis filter 110 in eight sets per frame. In this process, LSF can be extracted as one set per frame.

Meanwhile, a pitch is measured 140 from a residual component that has passed through the LP analysis filter 110 and a pitch is interpolated 150. For example, a characteristic waveform at eight pitches per frame, CW) may be extracted 160. In this process, the pitch can be extracted at a rate of 1 per frame.

The waveform is then sorted (170) from the extracted characteristic waveform (CW) and normalized (190) by calculating 180 power so that power at a rate of, for example, 8 per frame And the characteristic waveform normalized at the same rate can be extracted.

Through this process, speech parameters including the pitch of the speech, the characteristic waveform (CW), the power and the line spectrum frequency (LSF) can be extracted.

In step S550, the user-customized voice correction apparatus may change the first characteristics of a plurality of specific voices preset for the extracted voice parameters to the second characteristics having a predetermined normal voice. As shown in FIGS. 2 and 3, in this step, the voice parameters extracted in step S530 may be changed (300). For example, the pitch of the specific voice can be stabilized within a predetermined range. In this case, the voice can be corrected by stabilizing the unstable fundamental frequency of the hoarse voice, the rough voice, and the like.

In addition, for example, the spectral envelope can be reconstructed by correcting the line spectrum pairs (LSP) of the specific speech. In this case, the specific sound of the nose can be corrected. That is, the nasal nose affects the spectral formant of the voice, and a nasal formant, which is a new formant, is added due to the nasal cavity, or energy of a specific frequency band is absorbed Anti-formant may occur. This formant can be corrected by reconstructing the spectral envelope by correcting the LSP.

In addition, for example, the characteristic waveform of the specific sound can be separated into a SEW (smoothly evolving waveform) component and a REW (rapid evolving waveform) component, and the separated SEW component and REW component can be adjusted within a predetermined range. A characteristic waveform (CW) representing a residual signal of a speech is divided into a SEW component having a voiced sound characteristic and a REW component having an unvoiced sound characteristic, and by changing the respective SEW component and REW component, The characteristics of the unvoiced sound can be adjusted, and the tone color can be changed accordingly. The reason why the characteristic waveform (CW) is separated into the SEW component and the REW component is that the voice can be grasped by separating into voiced and unvoiced sounds. Generally, a voiced sound has a periodic characteristic of a signal waveform, and unvoiced sounds have a noise-like shape without any feature. Therefore, when the characteristic waveform CW passes through the low-pass filter, the SEW component is extracted, which represents the voiced component of the speech. In addition, when the above SEW component is removed from the characteristic waveform (CW), the REW component appears, which represents the unvoiced component of the speech. In this way, the characteristic waveform (CW) can be separated and the voiced and unvoiced components can be individually adjusted. Considering this with the pitch, pitch represents the period of the voice signal, so it appears only in the voiced part and not in the unvoiced part. Therefore, the SEW component and the REW component can be used as a criterion for determining whether the sound is unvoiced or not, and can be used as a pre-discrimination step for pitch stabilization.

On the other hand, the step S550 may include the step of changing the user-customized voice correction apparatus from the preset first tone range to the predetermined second tone range for the extracted voice parameters. At this time, either one of the first tone color range and the second tone color range may be a tone color range with a cold feeling, and the other may be a tone color range with a warm tone. Thus, by analyzing the parameters of the voice through a previously stored voice database and using the characteristics of the cold or warm feeling voices classified through the listening evaluation, the voice of the cold voice data The parameter may be changed to a warm-feeling parameter, or the voice parameter of the warm-feeling raw voice data may be changed to a cold-feeling parameter.

In step S570, the corrected voice data can be synthesized based on the voice parameters having the changed second characteristic. As shown in FIG. 3, the voice parameters changed according to the setting of the user may be synthesized through the WI synthesizer and output as the corrected voice. The process of synthesizing the corrected voice data from the changed voice parameters will be described in more detail with reference to FIG.

FIG. 5 is a block diagram illustrating a process performed in the WI synthesizer of a user-customized voice correction method for converting a tone color by parameter change according to an embodiment of the present invention. As shown in FIG. 5, power is denormalized (210) using normalized characteristic waveforms (CW) and power provided at a rate of 8 per frame, for example, The instantaneous pitch and CW may be generated 230 after the pitch provided in the pitch is interpolated 240 and the waveform is rearranged 220 at a rate of 1 per subframe therefrom. The phase track is then measured 250 at a rate of 1 per sample, and the 2D-to-1D transformation 260 may be performed based on the phase track value and the instantaneous CW. The thus transformed residual signal is provided to the LP synthesis filter 280, for example, at a rate of 160 per frame, and the LP synthesis filter may use the modified LSF interpolated value 270 together to reconstruct the corrected speech have. As described above, in the speech correction method according to the embodiment of the present invention, since the speech parameters are synthesized without being quantized between the WI analyzer and the WI synthesizer, the degradation of speech can also be reduced.

The present invention may be embodied in many other specific forms without departing from the spirit or essential characteristics of the invention.

100: WI analyzer 110: LP analysis filter
120: LSF interpolation 130: LPC analysis
140: pitch measurement 150: pitch interpolation
160: CW extraction 170: Waveform alignment
180: POWER calculation 190: POWER normalization
200: WI synthesizer 210: POWER denormalization
220: Rearrange waveform 230: Instantaneous pitch and CW generation
240: Pitch interpolation 250: Phase track measurement
260: 2D-to-1D conversion 270: LSF interpolation
280: LP synthesis filter 300: Parameter change
S510: receiving raw voice data
S530: extracting speech parameters including pitch, characteristic waveform (CW), power and line spectrum frequency (LSF) from the inputted original speech data;
S550: changing, for the extracted speech parameters, the first features of a plurality of predetermined specific sounds to the second features having a predetermined normal speech
S570: synthesizing the corrected voice data based on the voice parameters having the changed second characteristic

Claims (8)

A user-
(1) receiving raw voice data;
(2) extracting speech parameters including pitch, characteristic waveform (CW), power and line spectrum frequency (LSF) from the input original speech data;
(3) changing, with respect to the extracted speech parameters, first characteristics of a plurality of predetermined specific sounds to second features having a predetermined normal speech; And
(4) synthesizing the corrected speech data based on the speech parameters having the modified second characteristic,
In the step (3)
Wherein the specific voice comprises at least one selected from the group consisting of hoarse voice, rough voice, breathy voice and nasal voice,
In the step (3)
A change to the second features of the first features,
And reconstructing a spectral envelope by correcting a pair of line spectra (LSP) of the specific speech. Alternatively, the characteristic waveform of the specific speech may be classified into a smoothly evolving waveform (SEW) component and a rapidly evolving waveform (REW) And adjusting the separated SEW component and the REW component within a predetermined range. 2. The method of claim 1, further comprising:
delete delete delete delete The method according to claim 1,
Between steps (1) to (4) above,
Characterized in that the step of quantizing the original speech data does not include the step of quantizing the original speech data.
2. The method according to claim 1, wherein in the step (3)
(3-1) changing the extracted voice parameters from a preset first tone color range to a preset second tone color range, wherein the user-customized voice Correction method.
A user-customized voice correction apparatus for implementing a user-customized voice correction method for converting a tone color by changing a parameter according to any one of claims 1, 6, and 7.

Images