JP2012194389A - Falsetto detection device and singing evaluation device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce failure in detection when detecting a falsetto from singer's singing voice.SOLUTION: When generating voice information data including a harmonic tone rate, a pitch, and a lapse time as information from user's singing voice data (step S110), a control unit 10 allocates the voice information data to a voice distribution table (step S112). The control unit 10 subjects the voice information data to processing using a filter to successively calculate a calculation value (step S114). When the calculated calculation value exceeds a preliminarily determined threshold (YES; in step S116), the control unit 10 decides a position of the filter at timing when the calculated value exceeds the threshold (step S120). The control unit 10 detects voice information data included in a positively weighted region in the filter, as voice information data based on a falsetto (step S122).

Description

  The present invention relates to a technique for detecting a back voice from voice sung by a singer.

  Some karaoke apparatuses have a function of scoring the skill of singing by a singer. In this scoring, the karaoke device records the voice of the singer's singing, analyzes the recorded voice and detects the characteristics, and calculates the score by comparing the analysis result with the evaluation standard.

  By the way, when singing, a so-called back voice may be used. A reverse voice is a voice in a state of being turned over beyond the calling point. Since some songs make full use of the back voice when singing, if the skill of singing with the back voice can be reflected in the scoring, the scoring accuracy can be further improved. For example, Patent Document 1 discloses a method of detecting a falset in a singer's song. In the technique described in Patent Document 1, it is noted that the ratio of the harmonic component contained in the back voice is extremely small compared to the local voice, and the spectral characteristics in the singing voice by the singer are in a predetermined state. The voice of singing is recognized as a back voice when making a sudden transition.

JP 2007-310204 A

  However, when the local voice and the back voice are compared, the relationship between the proportions of the harmonic components in each of them varies depending on the singer. In other words, the back voice of a singer is determined by the relationship with the singing voice of the singer, and is not fixed for all singers. In the technique described in Patent Document 1, since a ratio in which the ratio of the harmonic component is small is determined in advance, and this is used as a common reference for back voice recognition to any singer, Even if singing in a back voice, the spectrum characteristic does not correspond to a predetermined state, and the back voice may not be detected. In this way, if the singer actually sings with a back voice but sometimes it is not detected as a back voice, the karaoke device cannot score the skill of the back singing well, and the result of the scoring is Dissatisfaction can be significant.

  This invention is made | formed in view of the above-mentioned background, and it aims at reducing a detection omission when detecting a back voice from the audio | voice of the song by a singer.

  In order to solve the above-described problem, the present invention provides a voice data acquisition unit that acquires voice data representing a voice when a singer sings, a ratio of a harmonic frequency to a fundamental frequency as a harmonic ratio, and the voice Based on the voice data acquired by the data acquisition means, the calculation means for periodically calculating the harmonic ratio in the voice represented by the voice data and the pitch of the voice according to the passage of time in the singing, and the harmonic ratio In the coordinate system composed of the first axis representing the pitch and the second axis representing the pitch, the coordinates corresponding to each harmonic ratio and pitch calculated by the calculation means are set to the combination of the harmonic ratio and the pitch. Allocating means for allocating each of the groups, and among the areas to which the plurality of sets are allocated by the allocating means, Audio data representing the singing voice of a portion corresponding to said set and provides a falsetto detecting apparatus characterized by comprising a falsetto detection means for detecting as the voice data representing the falsetto.

  In a preferred aspect, the back voice detection means is a filter applied to the coordinate system, and a positive region that is a region having a positive weight, and a side having a higher harmonic ratio than the positive region in the coordinate system or the A filter having a minus region that is a region having a negative weight and a lower pitch than the plus region, and a harmonic ratio lower than a first reference value that is predetermined in the coordinate system and predetermined. A filter moving means for moving the filter within a range in which the pitch is higher than the second reference value, and the filter moving means moves the filter every time the filter is moved by the filter moving means. A negative calculated value obtained by negatively weighting the number of pairs and a positive value for the number of pairs included in the plus region Adding means for adding a positive calculated value obtained by performing the identification, and specifying the position of the filter in the coordinate system when the addition result of the adding means exceeds a predetermined threshold value. The plus area of the filter at the position is the partial area, and the audio data corresponding to the set assigned to the partial area is detected as audio data representing a reverse voice.

  In another preferable aspect, the back voice detection means includes distribution specifying means for specifying a distribution of pitches included in the set for each overtone ratio in the first axis, and the distribution specifying means uses the harmonic specification ratio. Based on the distribution of pitches specified for each, from the range of overtone ratio when the maximum point of distribution appears at a pitch higher than a predetermined reference value, the voice data corresponding to the set is It is detected as voice data to represent.

  Moreover, this invention is reference audio | voice data showing each component sound which comprises the back voice detection apparatus of any one of Claims 1-3, and the music used as a singing object, Comprising: Reference voice data acquisition means for acquiring reference voice data in which a back voice flag is attached to a constituent sound to be generated in a back voice, each voice represented by voice data acquired by the voice data acquisition means of the back voice detection device, and the reference Corresponding to each constituent sound represented by the reference voice data acquired by the voice data acquisition means, the evaluation means for evaluating the singing by the singer according to the result of comparing the corresponding sounds, the back voice When the voice data representing the back voice detected by the back voice detection means of the back voice detection device corresponds to the reference voice data to which the flag is attached, the voice data representing the back voice is supported. Data may be provided as a singing evaluation device, characterized in that it comprises an evaluation means for performing a high evaluation as compared with the case that does not correspond.

  ADVANTAGE OF THE INVENTION According to this invention, it becomes possible to reduce a detection omission when detecting a back voice from the audio | voice of the song by a singer.

Block diagram showing hardware configuration of karaoke equipment Process flow diagram when back face detection processing is performed Diagram for explaining formula for calculating harmonic ratio The figure which shows the distribution table of the voice information data based on the local voice and the back voice The figure showing the filter applied to the voice distribution table in FIG. Figure with a filter applied to the distribution table of voice information data based on the local voice and back voice Diagram showing correspondence between detected back voice and guide melody Diagram showing the judgment result Block diagram showing the functional configuration of the control unit The figure showing the distribution table of the voice information data based on the local voice and the back voice concerning the modification 2 The figure showing the distribution table of the voice information data based on the local voice and the back voice concerning the modification 2 The figure showing the filter applied to the audio | voice distribution table which concerns on the modification 3. The figure showing the filter applied to the audio | voice distribution table which concerns on the modification 3. Process flow diagram when back face detection processing according to modification 4 is performed The figure showing the distribution number acquisition reference line which concerns on the modification 4. The figure showing distribution of voice information data concerning modification 4 The figure showing distribution of voice information data concerning modification 4 The figure showing distribution of voice information data concerning modification 4 The figure showing distribution of voice information data concerning modification 4 The figure showing distribution of voice information data concerning modification 4 The figure explaining the process which determines the area | region of the back voice which concerns on the modification 4.

Hereinafter, an embodiment of the present invention will be described.
<Embodiment>
<Configuration>
FIG. 1 is a block diagram illustrating a hardware configuration of the karaoke apparatus 100.
The karaoke apparatus 100 performs scoring on the user's singing. In particular, the karaoke device 100 detects a part sung in the user's singing voice and includes it as a scoring target for scoring. In this karaoke apparatus 100, a deduction method is adopted as a scoring method. Here, the deduction method starts from a perfect score when a user starts singing a certain karaoke piece (100 points if the score is 100), and is deducted when the user's singing does not satisfy the evaluation criteria. It is a method. As shown in FIG. 1, the karaoke apparatus 100 includes a control unit 10, a storage unit 20, an operation unit 30, a display unit 40, a communication control unit 50, an audio processing unit 60, a microphone 61, and a speaker 62. Are connected via a bus 70. The control unit 10 includes a CPU (Central Processing Unit), a RAM (Random Access Memory), a ROM (Read Only Memory), and the like. In the control unit 10, the CPU controls each unit of the karaoke apparatus 100 by reading out a computer program stored in the ROM or the storage unit 20, loading it into the RAM, and executing it. The control unit 10 has a time measuring function for measuring time.

  The operation unit 30 includes various operators and outputs an operation signal representing the content of an operation performed by the user to the control unit 10. The display unit 40 includes, for example, a liquid crystal panel, and displays lyrics telops, background images, and the like corresponding to each karaoke piece under the control of the control unit 10. The communication control unit 50 controls data communication between the karaoke apparatus 100 and a server apparatus (not shown) via a network (not shown).

  The microphone 61 outputs an analog audio signal representing the collected audio to the audio processing unit 60. The audio processing unit 60 includes an A / D (Analog / Digital) converter. When the analog audio signal output from the microphone 61 is converted into digital audio data and output to the control unit 10, the control unit 10 To get. In this way, the control unit 10 functions as a voice acquisition unit that acquires voice data representing a voice when a user (singer) sings. The audio processing unit 60 includes a D / A (Digital / Analog) converter, converts digital audio data received from the control unit 10 into an analog audio signal, and outputs the analog audio signal to the speaker 62. The speaker 62 emits a sound based on the analog audio signal received from the audio processing unit 60. The storage unit 20 is a storage unit for storing various data, and is, for example, an HDD or a nonvolatile memory. The storage unit 20 includes a plurality of storage areas such as an accompaniment data storage area 21, a video data storage area 22, a GM (Guide Melody) data storage area 23, and a user singing voice data storage area 24.

  The accompaniment data storage area 21 stores information related to accompaniment data representing accompaniment sound in each music piece. In the accompaniment data storage area 21, a “song number” that is a number for uniquely identifying a song, a “song name” that represents the name of each song, a “singer name” that represents the name of each song singer, A plurality of accompaniment data records including a plurality of items such as “file storage location” that is a storage location of a data file that is accompaniment data itself are stored. The accompaniment data file is, for example, a MIDI (Musical Instrument Digital Interface) format file.

  The video data storage area 22 stores the above-mentioned song number, lyrics data indicating the lyrics of each song, and background video data representing the background video displayed on the background of the lyrics in association with each other. The lyrics shown by the lyrics data are displayed on the display unit 40 as lyrics telop as the music progresses during karaoke singing. In addition, the background video represented by the background video data is displayed on the display unit 40 as the background of the lyrics telop as the music progresses during karaoke singing. In the GM data storage area 23, the above-described song number and data indicating the melody of the vocal part of the song, that is, guide melody data (hereinafter referred to as GM data) which is data specifying the content of the constituent sound (note) to be sung. ) Are stored in association with each other. The GM data serves as a reference for comparison when the control unit 10 evaluates the skill of singing by the user. The GM data has a back flag indicating whether or not each note should be sung back. For example, when the singer himself sings a certain note in the song “AAA”, the note flag is attached to the note in the state of “ON” in the GM data. On the other hand, the note sung by the singer himself is marked with the back-sound flag “OFF” for this note in the GM data. The GM data is described in the MIDI format, for example. Here, the local voice is the voice when speaking normally. The voice of a song sung with a local voice is rich in overtones (that is, there are many harmonic components in frequency). On the other hand, a back voice is a voice that has turned over from the local voice (beyond the screaming point). The voice of the song sung in the back voice has less harmonic components than the ground voice, but has a high pitch (pitch).

  In the user singing voice data storage area 24, the user's singing voice collected by the microphone 61 during the period in which the accompaniment data is being reproduced for each piece of karaoke music is converted into digital data by the voice processing unit 60. Audio data generated by the conversion is stored. This voice data is called user singing voice data. This user singing voice data is stored as a data file in a WAVE (RIFF waveform Audio Format) format, for example. The user singing voice data for each song is associated with the GM data of the song by the control unit 10.

<Operation>
Next, a method for detecting a back voice by the control unit 10 will be described with reference to FIGS.
FIG. 2 is a process flow diagram when the back voice detection process is performed. When the music is reserved by the user via the operation unit 30 (step S100; Yes), the control unit 10 searches for the reserved music from the storage unit 20 (step S102). Specifically, in step S102, the control unit 10 uses the song number of the selected song from each of the accompaniment data storage area 21, the video data storage area 22, and the GM data storage area 23 as a key, and relates to the song. Data is searched, and search result data is read into the RAM. After step S102, the control unit 10 reproduces the music based on the accompaniment data, video data, and GM data stored in the RAM (step S104). Specifically, in step S104, the control unit 10 causes the speaker 62 to emit sound based on the accompaniment data and the GM data, and causes the display unit 40 to display a video based on the video data. Then, the control unit 10 converts the user singing voice data obtained by converting the user singing voice collected by the microphone 61 into digital data by the voice processing unit 60 during the reproduction period of the music. The data is stored in the data storage area 24 (step S106).

  Next, the control part 10 calculates what becomes a harmonic ratio which represents the ratio of a harmonic component from user song voice data (step S108). First, basic words are explained, and then the harmonic ratio is explained. The “basic frequency” means the frequency of the lowest frequency component when a signal based on the sound of a certain note is expressed by synthesis of a sine wave. In addition, the component that is the pitch of this note is called “fundamental tone”. The “overtone” is a sound component having a frequency that is an integer multiple of 2 or more with respect to the frequency of the fundamental tone. From the above, when the vertical axis represents the power of the frequency component and the horizontal axis represents a biaxial coordinate system representing the frequency, the harmonic ratio can be captured as follows. “Power of fundamental frequency” is defined as “the area of the power of the frequency component of the fundamental frequency corresponding to the width of the frequency from the start of the peak to the end of the peak centered on the peak of the fundamental frequency”. Is the power of the frequency component of the overtone frequency whose center is the peak of the frequency 2 to n times the fundamental frequency and whose width corresponds to the frequency width from the start of the peak to the end of the peak. If defined as “total area of”, the harmonic ratio is represented by “power of frequency of harmonics / power of fundamental frequency”. In other words, the overtone ratio can be said to be the ratio of the overtone frequency to the fundamental frequency.

  The reason why the control unit 10 calculates the overtone ratio when detecting the back voice is as follows. As described above, the back voice has fewer harmonic components than the ground voice, but has a high pitch (pitch). Therefore, when considering a two-axis coordinate system in which the vertical axis is the harmonic ratio and the horizontal axis is the pitch, it is considered that the back voice is included more in the region where the harmonic ratio is low and the pitch is high. For this reason, the control unit 10 calculates the overtone ratio and uses it for detection of the back voice.

  In step S <b> 108, the control unit 10 calculates a harmonic overtone ratio from the user singing voice data at a predetermined sampling cycle such as 100 msec (milliseconds) in accordance with the passage of time in the user singing voice data. Moreover, the control part 10 acquires the elapsed time from the time of the start of karaoke music at the time of calculating a harmonic overtone ratio with the said sampling period by the time measuring function with which it was equipped. The control unit 10 stores, in the RAM, voice information data that is a combination of the elapsed time, the calculated overtone ratio, and the pitch of the user singing voice data at the time of calculating the overtone ratio (step S110). . Here, in the audio information data, one set including one elapsed time, one harmonic ratio, and one pitch is defined as one unit. Since the calculation of the harmonic overtone ratio is performed in units of sampling cycles, a plurality of pieces of audio information data having the same number as the total number of samplings included in the music reproduction period are eventually stored in the RAM of the control unit 10. Become. In the following, a user who utters (that is, sings) the voice that is the creator of the voice information data is referred to as the owner of the voice information data. Moreover, the audio | voice information data produced | generated in the above procedures from the user song audio | voice data showing an audio | voice are called audio | voice information data based on an audio | voice. The control unit 10 repeats the above process at the sampling period, thereby calculating the overtone ratio for the user singing voice data for the entire period of music reproduction and storing the voice information data in the RAM.

FIG. 3 is a diagram for explaining a formula for calculating the overtone ratio. In FIG. 3, the vertical axis represents the power of the frequency component, and the power increases as it progresses from bottom to top in FIG. 3. The horizontal axis represents the frequency, and the frequency increases as it proceeds from left to right in FIG. Area A1 is centered on the peak of the fundamental frequency, centered on the peak of the fundamental frequency, and the width of the power of the frequency component of the fundamental frequency corresponding to the width of the frequency from the start of the peak to the end of the peak, That is, it represents the power of the fundamental frequency described above. Further, the region A2 and the region A3 have frequency components of overtone frequencies whose center is a frequency peak of 2 to n times the fundamental frequency and whose width corresponds to the frequency width from the start of the peak to the end of the peak. This represents the total power area, that is, the power of the above harmonic frequency. Therefore, as described above, since the harmonic ratio is expressed by “power of harmonic frequency / power of fundamental frequency”, the calculation formula of harmonic ratio is as shown in (a).
(A) Overtone ratio = (A2 + A3 +... + An) / A1

  Following step S110, the control unit 10 assigns the plurality of audio information data stored in the RAM to the audio distribution table (step S112). Here, the voice distribution table is a table based on a two-axis coordinate system in which the vertical axis represents the harmonic ratio and the horizontal axis represents the pitch, and is stored in the RAM. FIG. 4 is a diagram illustrating an example of a voice distribution table of voice information data based on the voices of the local voice and the back voice. In FIG. 4, the vertical axis represents that the harmonic overtone ratio increases as it progresses from bottom to top in FIG. In addition, the horizontal axis represents that the pitch increases as it proceeds from left to right in FIG.

  The local voice area a is an example of an area when the voice information data based on the voice of the singing by the vocalist of the singer is assigned to the voice distribution table. That is, when a point is plotted at a location corresponding to the harmonic ratio and pitch included in the speech information data corresponding to the local voice in the voice distribution table, the set of the points falls within this local voice region a. . The back voice area b is an example of an area when the voice information data based on the voice of the singing voice of the singer is assigned to the voice distribution table. In other words, when points are plotted at locations corresponding to harmonic ratios and pitches included in each piece of audio information data corresponding to the back-sound in the sound distribution table, the set of points falls within this back-sound region b. Comparing the local voice area a and the reverse voice area b, the local voice area a is distributed in a relatively low to medium level with respect to the pitch, and a harmonic ratio is from a low level to a high level. On the other hand, the back voice region b is distributed from a medium to a relatively high pitch with respect to the pitch, and from a low to a medium overtone ratio. The voice distribution table looks like this because, as mentioned above, the back voice has less harmonic components and a higher pitch than the ground voice, while the local voice has a lower pitch than the back voice. This is because there is no bias like a back-sound regarding the frequency.

  When the voice information data is assigned to the voice distribution table in step S112, the voice information data is distributed in two areas, the local voice area a and the back voice area b, as shown in FIG. However, in this state, since it is not possible to accurately detect which voice information data is based on the back voice, the control unit 10 performs the following process so that the voice information data can be detected with high accuracy. . After step S112, the control unit 10 performs filtering using a filter on the audio information data group assigned to the audio distribution table, and calculates a calculated value (step S114).

  FIG. 5 is a diagram showing the filter F applied to the sound distribution table in FIG. 4, and the meanings of the vertical axis and the horizontal axis are the same as those in FIG. 4. As shown in FIG. 5, the filter F is composed of a combination of four regions formed by rectangles. In the filter F, negative weights are assigned to the upper left, upper right, and lower left areas Fa, Fb, and Fc (referred to as negative areas), and a positive weight is assigned to the lower right area Fd (plus Called area). As shown in FIG. 5, the minus area is on the higher harmonic ratio side or the lower pitch side than the plus area. The control unit 10 multiplies the number of pieces of audio information data included in each area by the weight assigned to each area, and uses the sum of the multiplication results as a calculated value. As described above, the back voice has fewer harmonic components than the ground voice, but the pitch is high, so that there is a high possibility that the voice information data is distributed at the lower right position in the voice distribution table. Therefore, if the control unit 10 multiplies the number of audio information data existing in this region by a positive weight, it is possible to obtain a high calculated value for the back voice. This is the reason why only the lower right region Fd is positively weighted in the filter F.

FIG. 6 is a diagram in which the filter F is applied to the distribution table of the voice information data based on the local voice and the back voice. In FIG. 6, the meanings of the vertical axis and the horizontal axis are the same as those in FIG. Based on the contents of FIGS. 4 to 6, the calculation example of the calculated value in step S114 will be described as follows. The control unit 10 calculates a negative value obtained by negatively weighting the number of pairs included in the negative region of the filter F, and a positive value obtained by performing positive weighting on the number of pairs included in the positive region. The calculated value is added to calculate the total calculated value. For example, “−2” is assigned to the upper left area Fa of the filter F, “−1” is assigned to the upper right and lower left areas Fb and Fc, and “+3” is assigned to the lower right area Fd. Assume that weights have been assigned. Here, “2” in the upper left area, “2” in the upper right area, “4” in the lower left area, and “10” in the lower right area, Suppose each was included. At this time, the control unit 10 calculates a calculated value by the following equation (b).
(B) (−2 × 2) + (− 1 × 2) + (− 1 × 4) + (3 × 10) = 20

  Returning to FIG. 2, when the calculated value does not exceed the predetermined threshold (step S116; No), the control unit 10 moves the filter F (step S118). Then, the control unit 10 repeats the calculation of the calculated value (step S114) and the movement of the filter F (step S118) until Yes is obtained in step S116. The above threshold value in step S116 is a result of assigning a plurality of voice information data created from the voice of singing by an unspecified number of users to the voice distribution table at the time of designing the karaoke apparatus 100. It may be obtained experimentally so that it is possible.

  Returning to FIG. 6, description will be given when the control unit 10 moves the filter F. A rectangle represented by a broken line represents a movable range MR of the filter F. When the control unit 10 moves the filter F in step S118, the region where the pitch is low and the region where the harmonic ratio is high are excluded from the movement range MR. Specifically, when all the sound information data is assigned to the sound distribution table, the control unit 10 counts the number of sound information data in ascending order of the overtone ratio included in the sound information data. When the count value reaches a predetermined ratio with respect to the total number of audio information data, the control unit 10 sets the harmonic ratio included in the audio information data at this time in the harmonic ratio direction in the movement range MR. The upper limit of movement (first reference value). The lower limit of the movement in the pitch direction in the movement range MR is the lowest harmonic ratio among the harmonic ratios included in the audio information data. On the other hand, the upper limit and the lower limit of the movement of the filter F in the pitch direction in the movement range MR are determined by the control unit 10 as follows. That is, when all the sound information data is assigned to the sound distribution table, the control unit 10 counts the number of sound information data in descending order of the pitch included in the sound information data. When the count value reaches a predetermined ratio with respect to the total number of audio information data, the control unit 10 determines the pitch included in the audio information data at this time as the pitch direction in the movement range MR. The lower limit of the movement of the filter F to the second (second reference value). The upper limit of the movement of the filter F in the pitch direction in the movement range MR is the highest pitch among the pitches included in the voice information data. That is, the control unit 10 moves the filter within a range in which the harmonic overtone ratio is lower than the predetermined first reference value and the pitch is higher than the predetermined second reference value in the audio distribution table. It can also be said.

  The reason why the moving range MR is set in this manner is that, as described above, there is a low possibility that voice information data based on the back voice is assigned to a low pitch area and a high harmonic ratio area in the voice distribution table. is there. In addition, the said predetermined ratio distinguishes a local voice and a back voice from the result of having allocated the several audio | voice information data created from the audio | voice of the singing by an unspecified many users to an audio | voice distribution table at the time of the design of the karaoke apparatus 100. In addition, what is considered to be suitable for evaluating the voice can be obtained experimentally and stored in the ROM of the control unit.

  Here, the movement of the filter F in step S118 may be performed as follows. In the voice distribution table, when the upper left corner of the filter F is positioned at a certain value height in the harmonic ratio, the control unit 10 filters the predetermined width (for example, 50 cents) in the positive direction of the horizontal axis. Every time F is moved, a calculated value is calculated. When the right end of the filter F comes into contact with the right end of the moving range MR, the control unit 10 positions the left end of the filter F at the left end of the moving range MR and has a predetermined width in the negative direction of the overtone ratio (for example, power Move by one unit). The control unit 10 causes the filter F to repeat the movement of drawing such a trajectory until the calculated value exceeds the threshold value.

  When the calculated value exceeds the predetermined threshold value (step S116; Yes), the control unit 10 specifies the position of the filter F (referred to as a detection position) at the position when the calculated value exceeds the threshold value ( Step S120). The fact that the calculated value calculated by the filtering process using the filter F exceeds a predetermined threshold means that the calculated value is predetermined in the region Fd having only positive weight among the regions in the filter F. This means that there is a sufficient number of audio information data that exceeds the threshold. Further, as described above, the back voice has less harmonic components than the ground voice, but the pitch is high, so there is a high possibility that the voice information data is distributed at the lower right position in the voice distribution table. . Therefore, after step S120, the control unit 10 detects that the audio information data included in the positively weighted area (that is, the lower right area Fd) in the filter F is audio information data based on the reverse voice ( Step S122). In other words, the control unit 10 assigns to a part of the area where the sound information data that is a set of a plurality of harmonic ratios and pitches is allocated, and which has a relatively low harmonic ratio and a high pitch. The user singing voice data corresponding to the received voice information data is detected as user singing voice data representing a back voice. That is, the control unit 10 detects all the user singing voice data detected in this way as a back voice even if it is, for example, a “whispering voice” or a “whispering voice”. In other words, in the present invention, the meaning of the term “back voice” includes all the sounds detected as described above.

  After step S122, the control unit 10 associates the voice information data based on the back voice with the GM data (step S124). A specific association method is as follows. As described above, in the voice information data, the overtone ratio, the pitch of the user singing voice data at the time of calculating the overtone ratio, and the karaoke piece of music at the time of calculating the overtone ratio at a predetermined sampling cycle are included. The elapsed time from the start time is stored in association with each other. Using this, the control unit 10 acquires the elapsed time from the voice information data detected as being based on the back voice, and associates it with GM data at a timing corresponding to the acquired elapsed time.

  FIG. 7 is a diagram schematically showing the correspondence between the detected back voice and the guide melody. In FIG. 7, the horizontal axis represents time, and the time elapses from left to right in FIG. 7. The vertical axis represents the pitch, and the pitch increases as it progresses from bottom to top in FIG. One scale on the vertical axis means a pitch of 200 cents (all sounds). That is, for example, in FIG. 5, the scale located one scale above the scale corresponding to the pitch “A4” represents the pitch of “B4”. Further, a scale located one scale below the scale corresponding to the pitch “A4” represents the pitch of “G3”.

  In FIG. 7, areas GM1 to GM3 and GM5 to GM7 represent guide melodies having pitches based on GM data. For example, in the period shown in FIG. 7, the A4 pitch sound lasts for the period T1, the D4 pitch sound lasts for the period T2, and the G4 pitch sound continues for the T3 period. If it lasts only for a period, the silent state lasts for the period of T4. In FIG. 7, the guide melodies GM1, GM2, and GM4 represented by a lattice pattern are set to sing with the local voice (that is, the back voice flag is “OFF”), and the guide melody is represented by an oblique stripe pattern. It is assumed that the melody GM3, GM6, and GM7 are set to sing in reverse voice (that is, the reverse voice flag is “ON”). Moreover, the continuous line 300 represents the pitch of the audio | voice at the time of the singing by the user represented by the user song voice data mentioned above. This is hereinafter referred to as a user singing voice curve 300.

  Returning to FIG. 2, after step S124, the control unit 10 determines whether or not the user sang in the GM data in which the back voice flag is “ON” (step S126). For example, in FIG. 7, an excerpt of the voice distribution table is displayed below, and in the region Fd detected as a back voice, points corresponding to each voice information data are represented by black circles. Here, it is assumed that the guide melody at the timing corresponding to the elapsed time included in certain audio information data g is the guide melody GM3. Further, it is assumed that the guide melody at the timing corresponding to the elapsed time included in the audio information data h is the guide melody GM7. As described above, the voice information data and the guide melody are associated with each other in step S124.

  Since the back melody flag of the guide melody GM3 is “ON”, the control unit 10 determines that the user has sung in the back voice at the timing when the user should sing in the back voice. On the other hand, since the voice information data based on the back voice is not associated with the guide melody GM6 in which the back voice flag is “ON”, the control unit 10 determines the timing at which the user should sing in the back voice at the timing of the guide melody GM6. , It is determined that it was not sung with a back voice. The control unit 10 can use the determination result for scoring. For example, since the karaoke apparatus 100 employs a deduction method, in the above case, the control unit 10 deducts at the timing of the guide melody GM6. In other words, when the user sings with the back voice at the time when the user should sing with the back voice, the control unit 10 can be said to perform higher evaluation as compared with the case where the user does not sing with the back voice at the time when the user should sing with the back voice. The calculation method of the evaluation result at the time of deduction used by the control unit 10 uses various known techniques such as frequency analysis and volume analysis using FFT (Fast Fourier Transform) as a technique for analyzing the user singing voice. , The evaluation result may be calculated for a predetermined evaluation item, or simply, if the back voice flag is not sung in the back voice at the timing of “ON”, a deduction may be made by a predetermined point. .

  Following step S126, the control unit 10 displays the determination result on the display unit 40 (step S128). FIG. 8 is a diagram illustrating the determination result. As described in the example of FIG. 7, during the period of T3 and T7 in which the back voice flag is “ON”, since the singing by the user was a back voice, a mark “◯” is displayed. On the other hand, during the period T6 in which the back voice flag is “ON”, since the singing by the user was a natural voice, a mark “Δ” is displayed. The control part 10 displays the determination result about such a back voice on the display part 40 with a scoring result after the user's song is complete | finished.

  FIG. 9 is a block diagram illustrating a functional configuration of the control unit 10. As illustrated in FIG. 9, the control unit 10 functions as a voice data acquisition unit 11, a calculation unit 12, an allocation unit 13, and a back voice detection unit 14. The back voice detection unit 14 includes a filter 141, a filter moving unit 142, and an adding unit 143. The calculating means 12 periodically calculates the overtone ratio in the voice represented by the user singing voice data and the pitch of the voice according to the passage of time in the singing. In the coordinate system composed of the first axis representing the harmonic ratio and the second axis representing the pitch, the assigning means 13 assigns the harmonics to the coordinates corresponding to each harmonic ratio and pitch calculated by the calculating means 12. Assign a set of ratios and pitches. The back voice detection unit 14 selects audio data corresponding to the group assigned to a part of the region where the plurality of sets are assigned by the assigning unit 13 and has a relatively low harmonic ratio and high pitch. , Detected as voice data representing a back voice. The filter 141 is a filter applied to the coordinate system, and includes a plus region that is a region having a positive weight, and a side having a higher harmonic ratio than the plus region or a side having a lower pitch than the plus region in the coordinate system. And a negative area that is a negative weighting area. The filter moving means 142 moves the filter within a range in which the harmonic overtone ratio is lower than the predetermined first reference value and the pitch is higher than the predetermined second reference value in the coordinate system. Each time the filter is moved by the filter moving means, the adding means 143 calculates a negative calculated value obtained by negatively weighting the number of sets included in the minus area of the filter and the set included in the plus area. And a positive calculated value obtained by applying a positive weight to the number of.

  Thus, according to the present invention, it is possible to reduce detection omissions when detecting a back voice from the voice of a song performed by a singer. In addition, according to the present invention, since a method of comparing with a data group prepared in advance is not used to detect a reverse voice, it is not necessary to prepare the data group in advance, and it is created from a voice of a song sung by a user. Among the audio information data that has been generated, the audio information data created based on the back voice does not correspond to the data group, and as a result, the back voice to be detected cannot be detected. Further, in the present embodiment, the filtering process is performed using a filter composed of a plurality of regions each of which is given a positive or negative weight in the coordinate system composed of the two axes. Thus, since the filter is composed of a region having a negative weight and a region having a positive weight and a region, the voice information data based on the local voice in the voice distribution table can easily obtain a negative calculated value. The voice information data based on it is easy to obtain a positive calculated value. As a result, the voice information data based on the local voice is easily separated from the voice information data based on the back voice, and the voice information data based on the back voice leaks from detection by being erroneously recognized as the voice information data based on the local voice. Can be reduced.

<Modification>
The above embodiment can be modified as follows. In addition, you may implement the following modifications suitably combining.

<Modification 1>
The setting for the rectangular width of each region in the filter F is not fixed as described in the embodiment, and the control unit 10 may be able to correct the setting. The correction method may be as follows. For example, in the karaoke apparatus 100, it is possible to input which user sings along with the reservation of karaoke music via the operation unit 30. And as an example, the control part 10 produces audio | voice information data from the user song audio | voice data based on a user's "Mr. A" song. The control unit 10 creates audio information data each time a song input by the user “Mr. A” as a singer of karaoke singing is played.

  In this way, when the voice information data having the user “Mr. A” as the owner of the voice information data is created a plurality of times by the control unit 10, the voice information data of “Mr. A” approaches a more accurate one. To go. In other words, compared to the case where the user “Mr. A” sings only one piece of music that sings less in the back voice, the user “Mr. A” sings in a part of the back voice, or sings in the whole voice (so-called false set). ) Accumulation of voice information data created for music or the like increases the number of voice information data. In general, as the number of data increases, the statistical result approaches an accurate one, and thus the voice information data accumulated in this manner is closer to the one representing the voice characteristics of the user “Mr. A”. It can be said that

  Based on the audio information data created as described above, the control unit 10 corrects the setting of the rectangular width in each region in the filter F. For example, for the rectangular width of each region in the filter F, the control unit 10 performs correction to reduce or widen the width of the line segment represented by the solid line in FIG. As a result, in FIG. 5, the size of each of the rectangles forming the four regions changes. Here, the correction of the width between the line segments to be corrected by the control unit 10 is based on the accumulated audio information data, and the audio information data based on a larger number of back voices is positively weighted. May be included in the allocated area.

  In addition, the control part 10 continues performing the correction | amendment of the setting mentioned above until completion | finish of the karaoke song by a user. That is, each time a user sings a karaoke song for a long time and more voice information data is accumulated, the above setting becomes more accurate in accordance with the characteristics of the user's voice. In this way, the control unit 10 can detect the back voice with higher accuracy in accordance with the characteristics of the user's voice and can also know the scoring result according to the characteristics of the user's voice. become.

<Modification 2>
The control unit 10 is not limited to using the result of detecting the back voice for scoring the song by the user, but may be as follows. For example, in FIG. 4, the local voice area a and the back voice area b are not so far apart in the pitch direction. This means that the owner of the voice information data in FIG. 4 has few sounds with no voice between the local voice and the back voice. 10 and 11 are diagrams illustrating a distribution table of audio information data based on the local voice and the back voice according to the second modification. In FIG. 10, as apparent from the comparison with FIG. 4, there is a certain distance in the pitch direction between the local voice area a2 and the back voice area b2. This means that in the singing of the owner of the audio information data in FIG. 10, there is a certain amount of sound that does not produce a voice between the local voice and the back voice. That is, the shorter the distance between the local voice area a2 and the back voice area b2 in the pitch direction, the more the user who performed the singing can sing in a wider and less interrupted pitch range. It can be said that there is.

  On the other hand, in FIG. 11, as apparent from the comparison with FIG. 4, there is a region overlapping in the pitch direction in the local voice region a <b> 3 and the back voice region b <b> 3. This means that in the singing of the owner of the audio information data in FIG. 11, there is a certain amount of sound with a height that allows the user to separately use the local voice and the back voice. That is, in the pitch direction, the singing was performed as the distance between the local voice area a3 and the back voice area b3 becomes a negative value (the local voice area a3 and the back voice area b3 overlap in the pitch direction). It can be said that the user has a high skill to properly use the local voice and the back voice.

  Based on the above-described concept, when detecting the back voice in step S122, the control unit 10 may determine a skill to properly use the local voice and the back voice and display the determination result on the display unit 40. As a determination method, there are the following methods. In the voice distribution table, the control unit 10 recognizes the pitch of the voice information data located at the leftmost end in the local voice region a as the lowest pitch that can be sung with the local voice. Moreover, the control part 10 recognizes the pitch of the audio | voice information data located in the most right end in the geophone area | region a as the highest pitch which can be sung by a geophone in the audio | voice distribution table. In addition, the control unit 10 determines the pitch of the voice information data located at the leftmost position in the area (the area assigned with the positive weight at the detection position of the filter F) detected as the reverse voice in the voice distribution table. It is recognized as the lowest pitch that can be sung. Moreover, the control part 10 recognizes the pitch of the audio | voice information data located in the right end in the area | region detected as a back voice in the voice distribution table | surface as the highest pitch which can be sung by a back voice. Further, when the result of the authorization is “the highest pitch that can be sung with the local voice> the lowest pitch that can be sung with the back voice”, the owner of the audio information data It is determined that the singing can be performed in a range of a height that can be used separately between the local voice and the back voice.

  The control unit 10 causes the display unit 40 to display the determination result. As for the display method, for example, the control unit 10 may say, “You hear a cry from C3 to E4, and a cry from G4 to A5” or “You hear a cry from C3 to A4, G4 to C5. You can make the display unit 40 display a message such as “you can use the local voice and the back voice for the G4 to A4 range”. Or you may make it the control part 10 display on the display part 40 as an image which can distinguish the sound range which a user can sing by visual recognition with a local voice and a back voice, respectively. If it does in this way, it will become possible for a user to recognize the sound range which can be sung by each of a local voice and a back voice about a user's own singing, and the sound range which can use a ground voice and a back voice mutually. Thereby, when several users perform karaoke, the way of enjoying that each singing sound range can be known as a result of scoring becomes possible. In addition, for example, a user who wants to practice singing alone can know from the determination results the difference in the singing range between the real voice and the back voice and the range that can be used properly for each singing. It is also possible to practice singing.

<Modification 3>
In the embodiment, the filter F is a combination of four regions configured by rectangles, but the shape of the filter F is not limited to this. 12 and 13 are diagrams showing filters applied to the voice distribution table according to the third modification. In the filter F2 in the example of FIG. 12, in the voice distribution table, a region assigned with a positive weight has a pentagonal shape, and a region with a low pitch and a high harmonic ratio is missing as compared with FIG. Yes. Further, an area to which minus weight is assigned is provided outside the area to which the plus weight is assigned. In FIG. 12, the area to which negative weighting is assigned is divided into three, indicating that weighting is assigned to each of the divided areas. A region assigned with a negative weight may not be divided, and one weight may be assigned to the entire region assigned with a negative weight. For the movement of the filter F2, the same method as in the embodiment may be used. In this case, the midpoint of the diagonal line segment formed at the upper left corner of the filter F2 may be replaced with the upper left corner of the filter F in the embodiment.

  In the filter F3 in the example of FIG. 13, the area to which positive weighting is assigned in the voice distribution table has a triangular shape with a right angle at the lower right, that is, the apex having the highest pitch and the lowest harmonic ratio. . Further, an area to which minus weight is assigned is provided outside the area to which the plus weight is assigned. In this case, if the control part 10 calculates | requires the gravity center of the whole triangle which comprises the filter F3, it will memorize | store this in RAM. After that, the center of gravity is replaced with the upper left corner of the filter F in the embodiment, and the filter F3 may be moved.

  Also in the filters F2 and F3 according to the modified example 3, there are regions having a positive weight and regions having a negative weight. In short, in the voice distribution table represented by the two-axis coordinate system, it is provided so as to cover a region having a positive weight and a region having a positive weight with respect to a range in which voice information data based on the back voice is easily assigned. It is sufficient that the filter is constituted by two types of areas, that is, an area having a negative weight. At this time, the area having a negative weight may be arranged so that the possibility that the voice information data based on the back voice is distributed is small. As described above, even when the filter according to the third modification is used, the same effect as that of the embodiment can be obtained.

<Modification 4>
The method of detecting a back voice from the audio information data group is not limited to the method using a filter as in the embodiment. FIG. 14 is a process flow diagram when the reverse voice detection process according to the fourth modification is performed. In FIG. 14, until step S112, the processing performed by the control unit 10 is the same as that in FIG. After step S112, the control unit 10 uses a distribution number acquisition reference line, which is a reference line for acquiring the number of audio information data distributed in the pitch direction of the audio distribution table (referred to as the distribution number), as a harmonic ratio. The direction is moved (step S114b). Then, the control unit 10 generates a voice information data number distribution line representing the distribution of the number of voice information data in the pitch direction (step S116b).

  FIG. 15 is a diagram illustrating a distribution number acquisition reference line according to the fourth modification. 15, the vertical axis and the horizontal axis are the same as those in FIG. Distribution number acquisition reference lines L1 to L5 represented by broken lines are examples of the distribution number acquisition reference line. 16a to 16e are diagrams showing the distribution of the audio information data in the fourth modification. In FIG. 16A to FIG. 16E, the vertical axis represents the number of audio information data, and indicates that the number increases as it progresses from bottom to top. Further, the horizontal axis represents the pitch, and the pitch increases as it progresses from left to right in FIGS. 16a to 16e. The voice information data number distribution lines M1 to M5 in FIGS. 16a to 16e correspond to the distribution number acquisition reference lines L1 to L5 in FIG. 15, respectively. That is, the voice information data number distribution line M1 represents the number of voice information data in the pitch direction based on the distribution number acquisition reference line L1 in FIG.

  In step S114b of FIG. 14, the control unit 10 moves the distribution number acquisition reference line by a predetermined width (for example, one unit of power) from the lower harmonic ratio to the higher harmonic ratio. In step S116b, the control unit 10 acquires the distribution number, that is, the number of audio information data distributed in the pitch direction of the audio distribution table, according to the distribution number acquisition reference line located at the value of the overtone ratio. A voice information data number distribution line is generated based on the acquired number of voice information data. In this way, the control unit 10 functions as a distribution specifying unit that specifies the distribution of pitches included in the audio information data for each harmonic ratio in the harmonic ratio direction of the audio distribution table. As shown in FIG. 16A, the audio information data number distribution line M1 corresponding to the distribution number acquisition reference line L1 has one peak. As shown in FIG. 15, this peak represents the number of audio information data based on the back voice. The voice information data number distribution line M2 to the voice information data number distribution line M4 has two peaks. As shown in FIG. 15, these peaks represent the number of voice information data based on the local voice and the number of voice information data based on the back voice. Furthermore, the audio information data number distribution line M5 corresponding to the distribution number acquisition reference line L5 has one peak. As shown in FIG. 15, this peak represents the number of voice information data based on the local voice.

  From FIG. 15 and FIGS. 16a to 16e, the number of peaks in the number distribution line of the audio information data is changed from one (on the right side of the pitch direction in FIG. 16a) to two and again (one in the pitch direction in FIG. 16e). It can be read that the control unit 10 indicates that the voice information data based on the back voice is not acquired in accordance with the distribution number acquisition reference line. After step S116b in FIG. 14, the control unit 10 determines whether the number of peaks in the audio information data number distribution line has changed from two to one (step S118b). The control unit 10 repeats the processes of step S114b and step S116b while determining No in step S118b. When the control unit 10 determines that the number of peaks in the voice information data number distribution line has changed from two to one (step S118b; Yes), the control unit 10 determines a region where voice information data based on the back voice exists (step S120b).

  The determination method in step S120b is, for example, as follows. First, in step S116b, the control unit 10 displays the first voice information data number distribution line when the first peak appears at a position higher than a predetermined reference in the pitch direction of the voice information data number distribution line. The corresponding distribution number acquisition reference line in the overtone ratio direction (first value) is stored in the RAM. Here, the predetermined standard is that, when the karaoke apparatus 100 is designed, more users are obtained from the result of assigning a plurality of audio information data created from singing voices by an unspecified number of users to the audio distribution table. Therefore, it may be obtained experimentally so that the back voice can be detected. In step S118b, the control unit 10 causes the RAM to store the value (referred to as the second value) in the overtone ratio direction of the distribution number acquisition reference line corresponding to the immediately preceding audio information data number distribution line.

  Based on the first value, the second value, and the predetermined reference in the pitch direction, the control unit 10 determines a region where the voice information data based on the back voice exists (see FIG. Step S120b). FIG. 17 is a diagram for explaining processing for determining a back voice area according to the fourth modification. In FIG. 17, the vertical axis represents the harmonic ratio, and indicates that the harmonic ratio increases as it progresses from bottom to top in FIG. 17. Also, the horizontal axis represents the pitch, and the pitch increases as it proceeds from left to right in FIG. A distribution number acquisition reference line L101 represented by a broken line is a distribution number acquisition reference line having the first value as a harmonic ratio, and a distribution number acquisition reference line L102 is a distribution having the second value as a harmonic ratio. It is a number acquisition reference line. A solid line S is a predetermined reference in the pitch direction (hereinafter referred to as a pitch reference S). The local voice area a4 is an example of an area when the voice information data based on the voice of the singing by the vocalist of the singer is assigned to the voice distribution table. The back voice area b4 is an example of an area when the voice information data based on the voice of the singing voice of the singer is assigned to the voice distribution table.

  As shown in FIG. 17, a shout region b4 exists in a region surrounded by the distribution number acquisition reference line L101, the distribution number acquisition reference line L102, and the pitch reference S. In step S120b of FIG. 14, the control unit 10 is an area surrounded by the distribution number acquisition reference line L101, the distribution number acquisition reference line L102, and the pitch reference S. And decide. And the control part 10 detects that the audio | voice information data contained in the area | region determined by step S120b is audio | voice information data based on a back voice (step S122b). In other words, the control unit 10 is based on the pitch distribution specified for each harmonic ratio, and the harmonic overtone when the maximum point of the distribution appears at a pitch higher than a predetermined reference value (pitch reference S). From the ratio range, the audio data corresponding to the audio information data is detected as audio data representing a reverse voice. The subsequent processing is the same as in the embodiment. As described above, the method according to the modification 4 can also achieve the same effect as that of the embodiment. Further, in the fourth modification, a back voice is detected based on the peak position of the number of pieces of voice information data distributed in the pitch direction and the number of the peaks obtained while changing the value in the harmonic ratio direction in the voice distribution table. ing. As described above, in the fourth modification, an area including a large amount of audio information data sandwiched between values based on the peaks in the distribution of the number of audio information data is determined as an area for detecting a reverse voice. It is possible to reduce the leakage of the voice information data based on the detection.

<Modification 5>
In addition to the singing evaluation apparatus, the present invention can be understood as a method for realizing these and a program for causing a computer to realize a voice evaluation function. Such a program may be provided in the form of a recording medium such as an optical disk storing the program, or may be provided in the form of being downloaded to a computer via the Internet or the like and installed and used.

  DESCRIPTION OF SYMBOLS 10 ... Control part, 20 ... Storage part, 21 ... Accompaniment data storage area, 22 ... Image | video data storage area, 23 ... GM data storage area, 24 ... User song audio | voice data storage area, 30 ... Operation part, 40 ... Display part, DESCRIPTION OF SYMBOLS 50 ... Communication control part, 60 ... Audio | voice processing part, 61 ... Microphone, 62 ... Speaker, 70 ... Bus | bath, 100 ... Karaoke apparatus, 300 ... User singing voice curve, a-a4 ... Geophonic region, b-b4 ... Back-sound region , G, h ... voice information data, A1, fundamental frequency power, A2, A3, harmonic frequency power, F-F3, filter, Fa-Fd, region, GM1-GM3, GM5-GM7, guide melody, L1 ~ L5, L101, L102 ... distribution number acquisition reference line, M1 to M5 ... voice information data number distribution line, MR ... movement range, S ... pitch reference

Claims (4)

Voice data acquisition means for acquiring voice data representing voice when the singer sings;
The ratio of the harmonic frequency to the fundamental frequency is defined as a harmonic ratio, and based on the audio data acquired by the audio data acquisition means, the harmonic ratio in the audio represented by the audio data and the pitch of the audio are represented in the time of the song. Calculating means for periodically calculating according to the progress of
In the coordinate system composed of the first axis representing the harmonic ratio and the second axis representing the pitch, the harmonic ratio and the pitch are set to the coordinates corresponding to each harmonic ratio and pitch calculated by the calculation means. Assigning means for assigning each set of
Audio data representing the voice of the singing of the portion corresponding to the group assigned to a part of the region where the harmonic ratio is relatively low and the pitch is high among the regions to which the plurality of sets are assigned by the assigning means A voice detection device comprising: voice detection means for detecting voice data representing voice. The back voice detection means includes
A filter applied to the coordinate system, which is a plus region that is a region having a positive weight, and a higher harmonic ratio than the plus region or a lower pitch than the plus region in the coordinate system. A filter having a negative area that is a negative weighting area;
Filter moving means for moving the filter within a range in which the harmonics ratio is lower than the first reference value determined in advance in the coordinate system and higher than the second reference value determined in advance;
Each time the filter is moved by the filter moving means, a negative calculated value obtained by negatively weighting the number of sets included in the minus area of the filter, and the set included in the plus area. Adding means for adding a positive calculated value obtained by applying a positive weight to the number of
When the addition result of the adding means exceeds a predetermined threshold, the position of the filter in the coordinate system is specified, and the plus area of the filter at the position is defined as the partial area. The voice detection device according to claim 1, wherein voice data corresponding to the set assigned to the area is detected as voice data representing a voice. The back voice detection means includes
A distribution specifying means for specifying a distribution of pitches included in the set for each overtone ratio in the first axis;
Based on the pitch distribution specified for each harmonic ratio by the distribution specifying means, from the range of the harmonic ratio when the maximum point of distribution appears at a pitch higher than a predetermined reference value, it corresponds to the set The voice detection device according to claim 1, wherein the voice data to be detected is detected as voice data representing a voice. The back voice detection device according to any one of claims 1 to 3,
Reference audio data acquisition that is reference audio data that represents each component sound that constitutes a song to be sung, and in which the component sound that is pronounced in a back tone is attached to the component sound that is a back voice flag. Means,
The voices represented by the voice data acquired by the voice data acquisition unit of the back voice detection device correspond to the constituent sounds represented by the reference voice data acquired by the reference voice data acquisition unit, respectively. The voice representing the back voice detected by the back voice detection means of the back voice detection device in the evaluation voice means for evaluating the singing by the singer according to the result of the comparison, wherein the back voice flag is attached to the reference voice data. A singing evaluation apparatus comprising: an evaluation unit that performs a higher evaluation when the data corresponds to the voice data representing the reverse voice than when the data does not correspond.

Images