KR101121505B1 - Method for extracting non-vocal signal from stereo sound contents - Google Patents

Abstract

The present invention relates to a method of extracting a non-vocal signal from a stereo sound source by removing a singer's voice from a stereo sound source so that only stereo accompaniment music can be output. After converting the difference signal of the sound source signal on each channel into the frequency domain, the frequency components on the left and right channels are compared with the frequency components of the difference signal, and the vocals of the frequency components on the left and right channels are compared. We propose a method of restoring a non-vocal signal by removing a frequency component in which a component is embedded.

Description

Method for extracting non-vocal signal from stereo sound contents}

The present invention relates to a method for extracting accompaniment music by removing a singer's voice from a sound source consisting of accompaniment music and a singer's voice, and more preferably, only stereo accompaniment music can be output by removing the singer's voice from a stereo sound source. The present invention relates to a method of extracting a non-vocal signal from a stereo sound source.

In general, a stereo sound source (AR: ALL Recorded) output to a speaker of a karaoke device, a song half cycle, an MP3 player, or other audio device is an accompaniment music using a vocal component (Vocal recorded) and at least one or more instruments. Non-vocal component (MR: Music recorded).

Unlike a mono sound source, such a stereo sound source records different musical instruments in each of the left and right channels, and then sends a voice signal to both channels so that sound source listeners can enjoy the music while feeling a three-dimensional effect.

The record producer works in the studio to produce such a sound source, and processes the final result sound source through an offline medium such as a CD or a file form and sells the sound source through an offline distribution network or an online distribution network.

The consumer purchases these sound sources, which are sold through the online and offline distribution networks, by the record producer in various ways and enjoys them through their own audio equipment.

Recently, there has been a growing desire among consumers to remove only accompaniment music from the original sound source and listen to the singer's upbringing, or to remove the singer's voice and listen to only the accompaniment music. I'm going.

The present invention is proposed as a result of this research activity, and proposes a new method in which accompaniment music extracts only a non-vocal signal by removing a singer's voice from a stereo sound source.

The present invention relates to a method of extracting a non-vocal signal from a stereo sound source, and to converting a sound source loaded on both left and right channels, and a difference signal of a sound source signal loaded on each of the left and right channels, respectively, into a frequency domain. The present invention proposes a method for restoring a non-vocal signal by comparing a frequency component on both left and right channels with a frequency component of the difference signal and removing a frequency component inherent in a vocal component among the frequency components on the left and right channels. .

In the present invention, the above-described sound source signals on the left and right channels and the frequency domain transformation of the difference signal are performed by a general frequency conversion process such as a fast fourier transform (FFT), a discrete cosine transform (DCT), or the like.

The frequency component comparison between the left and right channels of the present invention and the frequency component of the difference signal determine whether each frequency component on the left and right channels is equal to or greater than a predetermined multiple of the difference signal frequency component. If each frequency component on the right channel is larger than a predetermined multiple of the difference signal frequency component, it is determined that the vocal signal is carried out, and the corresponding frequency component is removed. Otherwise, it is judged to be a non-vocal signal, which is accompaniment music. Extract

A preferred embodiment of the method for extracting a non-vocal signal from a stereo sound source according to the present invention is (a) receiving a first sound source mixed with a vocal signal and a first non-vocal signal through a first channel, and receiving the vocal signal and the second ratio. Receiving a second sound source mixed with a vocal signal through a second channel; (b) removing the vocal signal by generating a difference signal between the first and second sound sources; (c) obtaining respective frequency components of the first and second sound sources and difference signals; (d) comparing α times the energy of each frequency component of the first and second sound sources with the energy of the frequency component of the difference signal (a rational number of α> 1); (e) When the energy of each frequency component of the first and second sound sources is greater than α times the frequency component energy of the difference signal, the vocal signal is present in the frequency components of the corresponding first and second sound sources. And determining that the first and second vocal signals do not exist when they are smaller than α times the frequency component energy of the difference signal.

According to an embodiment of the present invention, when it is determined in step (e) that vocal signals exist in the frequency components of the first and second sound sources, the corresponding frequency components of the first and second sound sources are removed. And preserving the frequency component when it is determined that the vocal signal does not exist in the frequency components of the first and second sound sources.

An embodiment according to the present invention further includes converting the frequency components of the first and second sound sources into a time domain by performing frequency inverse transform processing.

When the method of extracting the non-vocal signal from the stereo sound source according to the present invention is applied to the stereo sound source, only the accompaniment music recorded by the record producer in the state in which the singer's training is removed can be enjoyed.

In addition, when the method proposed in the present invention is applied to a karaoke apparatus or the like, a song half-cycle producer has an advantage of reducing the high cost of the production of accompaniment music. You can sing realistically against the backdrop of accompaniment music.

The technical idea of the present invention is to remove the vocal components by comparing the frequency components of the original sound signals of the left and right channels and the difference signals between the respective channels. Even if the technical scope of the present invention is included, it should be considered that the method of processing the non-vocal signal proposed in the present invention by simply adding and deleting the proposed method is also included in the technical scope of the present invention.

1 is a flowchart illustrating a method of extracting a non-vocal signal from a stereo sound source according to the present invention.
2A to 2F are diagrams illustrating a process in a frequency domain of a method for extracting a non-vocal signal from a stereo sound source according to the present invention.

Hereinafter, with reference to the drawings will be described in detail the technical spirit of the present invention.

1 is a process diagram illustrating a method of extracting a non-vocal signal from a stereo sound source according to the present invention.

As shown in step S110 of FIG. 1, the left channel 110 and the right channel 120 are channels through which a stereo sound source produced by a record producer is transmitted. The left channel 110 is a channel for transmitting accompaniment music (MR _L ) and the singer's voice (Vocal), and the right channel 120 is a channel for transmitting accompaniment music (MR _R ) and the singer's voice (Vocal).

Accompaniment music (MR _L , MR _R ) is the accompaniment music played by a certain instrument during the production of the record. In order to obtain stereo sound effects, accompaniment music of different instruments is generally embedded in each channel. Accompaniment music can be recorded equally in both the left and right channels. On the other hand, it is assumed in the present invention that the voice of the singer (Vocal) is the same recorded in both the left and right channels.

Meanwhile, in the present invention, for convenience of technical understanding and consistency of terminology, the accompaniment music (MR _L , MR _R ) is a non-vocal signal, the voice of the singer (Vocal) is a vocal signal, and the sound source is loaded on each channel. The accompaniment music and the voice of the singer will be represented by the first and second sound sources, respectively.

Therefore, in the present specification, the first sound source is a collective expression of accompaniment music (MR _L ) on the left channel and the voice of the singer (Vocal), and the second sound source is the accompaniment music (MR _R ) on the right channel and the voice of the singer (Vocal). It is desirable to understand) as a generic expression.

For reference, the first and second sound sources applied in the present invention mean a digital signal.

Therefore, in the case of an analog sound source, after being converted into a digital signal, the technical idea of the present invention described herein is equally applicable.

Hereinafter, the technical spirit of the present invention will be described in more detail with reference to the accompanying drawings.

As shown in step S120 of FIG. 1, the present invention performs the difference signal processing process 130 for the first and second sound sources transmitted through the left and right channels.

As a result of the difference signal processing, the vocal component, which is a component common to both channels, is removed, and only the difference signals MR _L -MR _{R of the} non-vocal signals MR _L and MR _R remain.

In a next step S130, Fast Fourier Transform (FFT) and DCT (Discrete) for the first sound source (MR _L + Vocal), the second sound source (MR _R + Vocal), and the difference signal (MR _L -MR _R ), respectively. Frequency components, such as Cosine Transform, are performed to extract frequency components.

By the FFT process, the first sound source MR _L + Vocal is converted into a frequency signal M _{L _ FFT} including both a non-vocal component MR _L and a frequency component of the vocal component Vocal, and a second sound source. (MR _R + Vocal) is converted into a frequency signal (M _{R _ FFT} ) including both the non-vocal component (MR _R ) and the frequency component of the vocal component (Vocal), and the difference signal (MR _L -MR _R ) is non- The vocal signal MR _L is converted into a frequency signal M _{D _ FFT} for a signal obtained by subtracting the non-vocal signal MR _R.

Next, a step of comparing the frequency energy for each of these signals is performed, which is expressed as follows.

[Equation 1]

Frequency bin energy of frequency signal (M _{L _ FFT} )> Frequency bin energy of α * difference signal (MR _L -MR _R )

[Equation 2]

Frequency bin energy of frequency signal (M _{R _ FFT} )> Frequency bin energy of α * difference signal (MR _L -MR _R )

Here, α is preferably a rational number greater than 1, and the range of α for the effective vocal signal extraction by experiment was approximately 1.5 <α <2.5. However, the range of α may deviate from the above range depending on the characteristics of the sound source originally recorded by the record producer, and a simple change of the α numerical range is naturally included in the technical idea of the present invention.

In Equation 1, when each frequency bin energy of the frequency signal M _{L _ FFT} is greater than each frequency bin energy of the corresponding α * difference signal MR _L -MR _R , the corresponding frequency signal (M _{L _ FFT} ) The frequency bin determines that the energy of the vocal signal is inherent and that the energy of the vocal signal is large and removes the frequency component of the corresponding frequency signal M _{L _ FFT} . That is, the frequency bin is initialized to zero. Performing this process individually for all frequency bins has the effect that the vocal signal is removed in the frequency component of the frequency signal M _{L _ FFT} .

Similarly, according to Equation 2, the same process as that of Equation 1, i.e., each frequency bin energy of the frequency signal M _{R _ FFT} corresponds to the α * difference signal MR _L -MR _R corresponding thereto. If it is larger than the energy, the corresponding frequency signal M _{R _ FFT} determines that the energy of the vocal signal is inherent and that the energy of the vocal signal is large, and thus the frequency of the corresponding frequency signal M _{L _ FFT} . Remove the ingredients. That is, the frequency bin is initialized to zero. Performing this process individually for all frequency bins has the effect that the vocal signal is removed in the frequency component of the frequency signal M _{R _ FFT} .

As can be seen in the next step (S160), the frequency bin that does not satisfy the conditions of equations 1 and 2 is left as the non-vocal signal component is predominant.

Therefore, when the processing of Equation 1 and Equation 2 is completed, the frequency bin component having a large vocal signal component is removed and the frequency bin component having a weak or no component of the vocal signal remains.

In the next step (S170), if the time domain transformation (Inverse FFT, Inverse DCT, etc.) is performed on the frequency components processed through the equations 1 and 2, the first sound source (MR _L + Vocal) and the second sound source (MR _R + It is possible to obtain an effect of removing the vocal signal (Vocal) from Vocal.

Therefore, when the method according to the present invention is implemented, there is an advantage that only the accompaniment music MR _{L which} is the non-vocal signal of the first sound source and the accompaniment music MR _{R which} is the non-vocal signal of the second sound source can be extracted and listened to.

However, even if the non-vocal signal extraction method from the stereo sound source proposed in the present invention is implemented, 100% of the accompaniment music recorded during the production of the recording cannot be extracted. However, as a result of the experiment, as a result of separately listening to the original accompaniment music and the final accompaniment music extracted by the implementation of the present invention, the difference in hearing could not be felt significantly, which is the final purpose of the technical idea to be implemented in the present invention. It also responded to.

2A to 2F are views for more visually explaining a method of extracting a non-vocal signal from a stereo sound source, which is a technical idea of the present invention.

2A is a conceptual diagram of a frequency spectrum of a first sound source, and FIG. 2B is a conceptual diagram of a frequency spectrum of a second sound source.

FIG. 2C is a conceptual diagram of the frequency spectrum of the difference signal between the first sound source and the second sound source shown in FIGS. 2A and 2B, and FIG. 2D is a conceptual diagram of the frequency spectrum multiplied by the frequency spectrum of FIG. 2C.

FIG. 2E is a frequency spectrum after removing a frequency bin determined to include vocal components by comparing the energy of the frequency spectrum of FIG. 2A with the frequency spectrum of FIG. 2D, and FIG. 2F is the frequency spectrum of FIG. 2B and the frequency spectrum of FIG. 2D. This is a conceptual diagram of the frequency spectrum after removing the frequency bins determined to include the vocal component by comparing the energies.

As shown in FIG. 2E and FIG. 2F, when the non-vocal signal extraction method according to the present invention is used, the main components of the non-vocal signal components included in the first and second sound sources may be extracted and converted into the time domain. As a result of listening in a real laboratory, the sound of the non-vocal signal included in the first and second sound sources and the frequency component of FIGS. Could know.

When the method of extracting a non-vocal signal from a stereo sound source according to the present invention is applied to various sound apparatuses including a computer, it is possible to efficiently enjoy the accompaniment music in which the singer's voice is removed.

Since the implementation of the present invention is generally programmed and processed in an acoustic device, it can be applied to various electronic media capable of downloading a program enabling the implementation of the present invention, such as a computer, a mobile phone, and an MP3 device, in addition to a half cycle. It's very simple and especially useful for smartphone applications.

Claims (3)

In the non-vocal signal extraction method from a stereo sound source,
(a) receiving a first sound source mixed with a vocal signal and a first non-vocal signal through a first channel, and receiving a second sound source mixed with the vocal signal and a second non-vocal signal through a second channel ;
(b) removing the vocal signal by generating a difference signal between the first and second sound sources;
(c) obtaining respective frequency components of the first and second sound sources and difference signals;
(d) comparing α times the energy of each frequency component of the first and second sound sources with the energy of the frequency component of the difference signal (a rational number of α>1);
(e) When the energy of each frequency component of the first and second sound sources is greater than α times the frequency component energy of the difference signal, the vocal signal is present in the frequency components of the corresponding first and second sound sources. And determining that the first and second vocal signals do not exist if they are smaller than α times the frequency component energy of the difference signal. The method of claim 1,
(f) In step (e), if it is determined that a vocal signal exists in the frequency components of the first and second sound sources, the corresponding frequency components of the first and second sound sources are removed, and the first and second sound sources are removed. And preserving the frequency component when it is determined that the vocal signal does not exist in the frequency component of the second sound source. The method of claim 2,
and (g) converting the frequency components of the first and second sound sources subjected to step (e) into a time domain by performing a frequency inverse conversion process.

Images