1377557 ‘ 第97]48492號專利說明書修IE本 修正日期：101年9月27日 六、發明說明： 【發明所屬之技術領威】 本發明主要有關於聲音信號分析與合成技術’特別係 有關於一種歌聲信號修正之裝置與方法。 【先前技術】 近年來，歌唱漸漸成為廣受歡迎的娛樂方式，然而坊 間所流行的包廂式卡拉OK (KTV) —般所配備的影音設 備雖然可提供歌詞與拍子提示、升降音調、以及迴音等音 效，但卻未針對荒腔走板的歌聲提供修正或美化之功能。 另一方面，在聲音分析與合成領域中，也有對於樂器 聲或歌聲合成，也就是通稱為樂音合成之研究，目前習知 的樂音合成方法大致上可分為時域（time domain)以及頻 域（frequency domain )兩類方法。 時域的樂音合成方法如：基頻同步累加法（pitch1377557 'Patent 97] No. 48492 Patent Specification Revision IE Revision Date: September 27, 101. VI. Description of the Invention: [Technology Leading to the Invention] The present invention mainly relates to sound signal analysis and synthesis technology. A device and method for correcting singing voice signals. [Prior Art] In recent years, singing has gradually become a popular entertainment method. However, the popular karaoke (KTV)-like audio and video equipment can provide lyrics and tempo, lifting tones, echoes, etc. Sound effects, but it does not provide correction or beautification for the songs of the desert. On the other hand, in the field of sound analysis and synthesis, there are also studies on instrumental or vocal synthesis, which is commonly referred to as musical tone synthesis. Currently, conventional music synthesis methods can be roughly divided into time domain and frequency domain. (frequency domain ) Two types of methods. Time domain music synthesis method such as: fundamental frequency synchronization accumulation method (pitch
Synchronous Overlap and Add，PSOLA )、基頻波形時間 比例内插法（time-proportionated interpolation of pitch waveforms，TIPW )、以及波表合成法（Wavetable Synthesis); —般來說，時域的歌聲合成方法事先需要的 參數分析較為簡單，且合成處理的計算量較少，然而，上 述的基頻同步累加法和基頻波形時間比例内插法係源自 於語音合成之技術，故不甚適用於歌聲的合成上，因為歌 聲的音長以及音高的變化幅度都較大；而波表合成法則是 一種用於樂器聲合成的方法，所以不適合使用於歌聲信號 的合成。 頻域的樂音合成方法主要可分為三類：（A )加法式 1377557 第97148492號專利說明書修正本 修正曰期：101年9月27日 合成（Additive Synthesis )，先產生各個譜波後再相加， 如：弦波模型（Sinusoidal Model )之樂音合成方法； 減法式合成（Subtractive Synthesis )，如：線性預測編碼 (Linear Predictive Coding )之合成方法；（C)頻率調變 式合成（Frequency Modulation) ’此方法多用於樂哭聲 之合成。 總體來說，上述之樂音合成方法除了針對樂器聲 理外，對於歌聲之處理則多以内建語料庫（C〇r{)Us) 成出特殊樂音或特殊效果之歌聲，並非針對演唱^ 來’ 直接進行修正或美化。 /、Θ 的歌聲 【發明内容】 根據本發明一實施例所述之一種歌聲修正裂置 一歌曲資料庫單元，儲存複數歌曲所對應之複數’ 訊；一發音錄製單元，錄製上述歌曲之一者對靡樂每資 信號’並將上述歌聲信號切割為複數音節（syl丨 歌聲 一歌聲k號分析單元，將上述各音節分別分割為勺）， 音框（frame)之一音框序列，再對上述各音框=括複數 聲學模型進行一既定聲學分析程序而取得^應二別以了 學模型之複數聲學參數；一聲學參數調整單元二；上述聲 樂譜資訊調整上述聲學參數以得到複數調整過上述 數；一歌聲信號合成單元，根據上述調整過的聲战，學參 上述聲學模型去產生複數合成音節；以及—歌聲二二數與 單元，串接上述合成的音節以產生—修正的歌^號串接 根據本發明一實施例所述之一種歌聲修正方广\。 取得一歌曲之樂譜資訊；錄製—演唱人所。曰 去，包括 曰处歌曲之複 4 1377557 第97M觀號專利說明書修正本 修正日期：]01年9月27曰 f音節’♦將上述各音節分別分割為包括複數音框之一 2列’，上逑各音框分別以一聲學模型進行一既定“ 於上述聲學模型之複數聲學參數，·根 聲㈣複數調整過的 ，予f數，根據上述調整過的聲學參數與上述聲學模 生出複數個合成的音節；以及串接、 生-修正的歌聲信號。 成的曰即以產 根據本發明另一實施例所述之一種歌聲修正方法，々 歌曲之樂譜資訊;錄製-演唱人所唱上述歌曲： ，聲w;將上述歌聲信號㈣為複數 ^ 即分別分割為包括複數音框之一立 $、各曰 庫於上过签t 進行一既定聲學分析程序而取得對 複數聲學參數；根據上述樂譜資㈣ 二數以得到複數調整過據; 的音節;以及串接上述合成數個合成 號。 成的曰即以產生一修正的歌聲信 法中揭露本二與優點，部份將於實施方 由實施過程中；法中明顯觀察得到、部份則可 利宣2由附屬項專 上述發明内容和接下央每 、Synchronous Overlap and Add (PSOLA), time-proportionated interpolation of pitch waveforms (TIPW), and Wavetable Synthesis; in general, time domain vocal synthesis methods The required parameter analysis is relatively simple, and the amount of computation of the synthesis process is small. However, the above-mentioned fundamental frequency synchronization accumulation method and the fundamental frequency waveform time proportional interpolation method are derived from the technique of speech synthesis, so it is not suitable for singing voice. In synthesis, because the pitch length of the singing voice and the pitch change are large, the wavetable synthesis method is a method for the sound synthesis of the instrument, so it is not suitable for the synthesis of the singing voice signal. The music sound synthesis methods in the frequency domain can be mainly divided into three categories: (A) Addition method 1377557 Patent specification 97148492 Revision of this revision period: Additive Synthesis on September 27, 101, first generate each spectrum wave and then phase Plus, such as: Sinusoidal Model music synthesis method; Subtractive Synthesis, such as: Linear Predictive Coding synthesis method; (C) Frequency Modulation Synthesis (Frequency Modulation) 'This method is mostly used for the synthesis of music crying. In general, the above-mentioned musical sound synthesis method is not only for the instrumental sounds, but also for the processing of the singing voice. The built-in corpus (C〇r{)Us) is used to produce a special musical tone or a special effect singing voice, not for the singing ^ to 'directly Make corrections or beautify. [Speech of the present invention] According to an embodiment of the present invention, a song sound correction splitting a song database unit stores a plurality of sounds corresponding to a plurality of songs; a sound recording unit for recording one of the songs For the music per-signal signal 'cut the above-mentioned singing voice signal into a complex syllable (syl sing-song-song-song k-number analysis unit, respectively divide the above-mentioned syllables into spoons), one of the frames of the sound box (frame), and then Each of the above sound frames includes a complex acoustic analysis program to obtain a complex acoustic analysis program to obtain a complex acoustic parameter of the learning model; an acoustic parameter adjusting unit 2; the above acoustic vocal information adjusts the acoustic parameters to obtain a complex adjustment The above-mentioned number; a singing voice signal synthesizing unit, according to the adjusted sound warfare, learning the above acoustic model to generate a plurality of synthesized syllables; and - a singing voice two-two number and unit, concatenating the synthesized syllables to generate a corrected song ^ The number is spliced in accordance with an embodiment of the present invention. Get the music information of a song; record - singer.曰 , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , Each of the upper voice frames is respectively subjected to an acoustic model to determine a plurality of acoustic parameters of the acoustic model, and the root sound (four) is adjusted in plural, and the f-number is obtained according to the adjusted acoustic parameters and the acoustic model. a synthesized syllable; and a concatenated, raw-corrected vocal signal. The vocal correction method according to another embodiment of the present invention produces a musical score information of the song; the recorded-singer sings the above-mentioned song : , sound w; the above-mentioned singing voice signal (four) is plural ^ is divided into a plurality of complex sound boxes, respectively, and each of the library is subjected to a predetermined acoustic analysis program to obtain a complex acoustic parameter; Capital (4) the second number to obtain the plural syllabic; and the concatenation of the above synthesis number of composite numbers. The resulting 曰 is to produce a modified vocal signal to reveal the second and the advantages, the Ministry Will embodiment of the process of embodiment; clearly observable process, part of the benefit can be declared by the dependent claims 2 designed to take over and above invention each central,
例舆說明，然其並非用以限定本;:方S 之保護範圍當視後附之申靖 6二圍’因此本發明 甲口月專利乾圍所界定者為準。 【實施方式】 1377557 修正日期：〗〇1年9月27曰 ‘ 第9714討92號專利說明書修正本 為使本發明之上述目的、特徵和優點能更明顯易懂， 下文特舉一些較佳實施例，並配合所附圖式，作詳細說明 如下： 第1圖係顯示根據本發明一實施例所述之歌聲修正裝 置。，此實施例中，一歌聲修正裝置2〇，具有一歌曲資 料庫單7G 21用以儲存複數歌曲所對應之複數樂譜資訊， 包括各音符之音高、音長和音量資訊。當歌聲修正裝置 20接收到來自一演唱人1〇所唱之對應於上述歌曲之一者 之歌聲仏號時，由一發音錄製單元22將上述歌聲信號錄 製起來，此錄製過程可為—次一個音節或一次連續錄製整首 歌曲的方式進行，若為一未連續錄製整首歌曲，則一發音錄 製單元22進一步將上述歌聲信號切割為複數音節。一歌聲 信號分析單元23將上述各音節分割為包括複數音框之一 音框序列’再對上述各音框分別以一聲學模型進行一既定 聲學分析程序而取得對應於上述聲學模型之複數聲學參 數’上述聲學杈型可為譜波加雜音模型（Harrnonic pius Noise Model ’ HNM)或相位語音編碼模型（phase Vocoder Model，PVM)，而上述之既定聲學分析程序可參閱參考 文獻[1] Stylianou ’ Yannis，「Harmonic plus Noise Models for Speech，Combined with StatisticalMethods，for Speech and Speaker Modification」，Ph.D. thesis，Ecole Nationale Superieure des Telecommunications 5 Paris 5 France 5 1996 > 以及參考文獻[2] Moore，F. R. ’ 「Elements of Computer Music」，Prentice-Hall ’ 1990。若歌聲信號分析單元23 所根據的聲學模型為諧波加雜音模型，則在分析後所得出 對應於上述音框之上述聲學參數之一者包括一組諧波參 6 1377557 . 第97148492號專利說明書修正本 修正日期：101年9月27日 數與一組雜音參數；若歌聲信號分析單元23所根據的聲 學模型為相位語音編碼模型，則在分析後所得出對應於上 述音框之聲學參數包括一組頻率參數與一組音強 (amplitude )參數。一聲學參數調整單元24根據歌曲資 料庫單元21所儲存之樂譜資訊調整上述聲學參數以調校 演唱人10所唱之歌聲，例如：調整上述聲學參數中有關 - 音高之參數以校正音調（pitch)不準確之問題、調整上述 聲學參數中有關音長之參數以校正拍子（tempo )不對的 問題、或甚至將上述聲學參數所對應之諧波基本頻率作一 頻率值之上下抖動以加入抖音（vibrato )之特性，最後得 到複數調整過的聲學參數；值得注意的是，為了讓演唱人 的音色（timbre )不變，在上述參數調整的過程中，必須 把上述聲學參數與上述調整過的聲學參數各別對應之頻 譜包絡曲線形狀保持相同。 接下來，若上述聲學信號作分析時採用的是諧波加雜 音模型，則一歌聲信號合成單元25進一步以一諧波加雜 音合成方法將上述調整過的聲學參數合成為複數個合成 的音節；若上述聲學信號作分析時採用的是相位語音編碼 模型，則歌聲信號合成單元25進一步以一相位語音編碼 合成方法將上述調整過的聲學參數合成為複數個合成的 音節；上述之諧波加雜音合成方法與上述相位語音編碼合 成方法於參考文獻[1] ' [2]中已有詳細說明，故在此不多 • 做贅述。最後，由一歌聲信號串接單元26，將上述合成 - 的音節串接成一修正的歌聲信號。 第2圖係顯示根據本發明一實施例所述之歌聲修正方 法之流程圖。該方法一開始接收到來自一演唱人所欲選唱 7 〇/ 粧_-年9肋日 吨)牛』於是從—歌曲資料庫令取得該歌曲之毕雄資 二，S2〇 ’上述樂譜資訊包括各音符之音^口 曰里。然後一個音節一個音節各 、 ==/(步將上述各音上 -聲學模型進行定t【23)再對上述各音框分別以 聲學模型之複數聲;參\聲=^^ 雜音模型或相位語音編碼模型，而 =序可㈣參考文獻[1]、[2]。若步驟S24所根攄 述立為譜波加雜音模型，則分析後所得出對庫於上 迷音框之聲學表數白杯— 1 ^ ^ 步驟幻4所根據的聲學_===組雜音參數；若 後所得出對岸於口立:曰編碼模型，則分析 與一組音強；X日框之聲學參數包括-組頻率參數 接下來，根據上述樂譜資訊調整 ，調整過的聲學參數（步驟S25)，調整=到 =述聲學參數中有關音高之參數；調 :二學參數中有關音長之參數以校正拍子ί 頻率值之±7抖所對應之基本頻率作— 了讓演唱人的音色“：性；值得注意的是，為 聲學參數與上述卜必須把上述 曲線形㈣軸同m學/數各卿叙頻譜包絡 的聲學參數與上述聲學模型:畢之後根據上述調整過 (步驟伽）。最^ =型去產生出複數個合成的音節 之歌聲信號(步=;)串接上述合成的音節以得到-修正 1377557 第97148492號專利說明書修正本 修正曰期·· H)1年9月27曰 第3圖係顯示根據本發明另一實施例所述之歌聲修正 方法之流程圖。方法一開始接收到來自一演唱人所欲選唱 之歌曲資訊，於是從一歌曲資料庫中取得該歌曲之樂譜資 訊（步驟S31)，上述樂譜資訊包括各音符之音高、音長 和音量。然後錄製演唱人所唱之整首歌的歌聲信號（步驟 S32)，將上述歌聲信號分割為複數個音節（步驟S33)，再 將上述各音節分別分割為包括複數音框之一音框序列（步 驟S34)，再對上述各音框分別以一聲學模型進行一既定 聲學分析程序而取得對應於上述聲學模型之複數聲學參 數（步驟S35)，上述聲學模型可為諧波加雜音模型或相 位語音編碼模型，而上述之既定聲學分析程序可參閱參考 文獻[1]、[2]。若步驟S35所根據的聲學模型為諧波加雜 音模型，則分析後所得出對應於上述音框之聲學參數包括 一組諧波參數與一組雜音參數；若步驟S35所根據的聲學 模型為相位語音編碼模型，則分析後所得出對應於上述音 框之聲學參數包括一組頻率參數與一組音強參數。 接下來，根據上述樂譜資訊調整上述聲學參數以得到 複數調整過的聲學參數（步驟S36)，調整之作法如：調整 上述聲學參數中有關音高之參數以校正音調不準確之問 題、調整上述聲學參數中有關音長之參數以校正拍子不對 的問題、或甚至將上述聲學參數所對應之基本頻率作一頻 率值之上下抖動以加入抖音之特性；值得注意的是，為了 讓演唱人的音色保持不變，在步驟S36中，必須把上述聲 學參數與上述調整過的聲學參數各別對應之頻譜包絡曲 線形狀保持相同。參數調整完畢之後，根據上述調整過的 聲學參數與上述聲學模型去產生出複數個合成的音節（步 9 1377557 • 第97148492號專利說明書修正本 修正日期：101年9月27曰 驟S37)。最後，串接上述合成的音節以得到一修正之歌 聲信號（步騾S38)。 本發明雖以範例揭露如上，然其並非用以限定本發明 的範圍，任何熟習此項技藝者，在不脫離本發明之精神和 範圍内，當可做些許的更動與潤飾，因此本發明之保護範 圍當視後附之申請專利範圍所界定者為準。 10 1377557 修正日期：101年9月27日 第97148492號專利說明書修正本 【圖式簡單說明】 第1圖係顯示根據本發明一實施例所述之歌聲修正裝 置。 第2圖係顯示根據本發明一實施例所述之歌聲修正方 法之流程圖。 第3圖係顯示根據本發明另一實施例所述之歌聲修正 方法之流程圖。 【主要元件符號說明】 10〜演唱人； 20〜歌聲修正裝置； 21〜歌曲資料庫單元； 22〜發音錄製單元； 23〜歌聲信號分析單元； 24〜聲學參數調整單元； 25〜歌聲信號合成單元； 26〜歌聲信號串接單元。For example, it is not intended to limit this; the scope of protection of Party S shall be deemed to be attached to Shenjing 6's circumference, so the definition of the patent of the present invention shall prevail. [Embodiment] 1377557 Revision Date: 〗 〖September 27 曰 第 14 14 14 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 97 For example, the following is a detailed description of the following: FIG. 1 shows a singing voice correcting apparatus according to an embodiment of the present invention. In this embodiment, a song sound correction device 2 has a song data library 7G 21 for storing the plurality of score information corresponding to the plurality of songs, including the pitch, length and volume information of each note. When the singing voice correcting device 20 receives the singing voice nickname corresponding to one of the songs sung by a singer, the vocal recording signal is recorded by a utterance recording unit 22, and the recording process can be one time. The syllable or the method of continuously recording the entire song is performed. If the entire song is not continuously recorded, the utterance recording unit 22 further cuts the vocal signal into a plurality of syllables. The vocal signal analysis unit 23 divides each of the syllables into a sequence of sound boxes including a plurality of syllables, and then performs a predetermined acoustic analysis procedure on each of the phonological frames by an acoustic model to obtain a plurality of acoustic parameters corresponding to the acoustic model. 'The above acoustic 杈 can be a Harvonic Pius Noise Model 'HNM' or a Phase Vocoder Model (PVM), and the above-mentioned established acoustic analysis program can be found in the reference [1] Stylianou ' Yannis , "Harmonic plus Noise Models for Speech, Combined with Statistical Methods, for Speech and Speaker Modification", Ph.D. thesis, Ecole Nationale Superieure des Telecommunications 5 Paris 5 France 5 1996 > and references [2] Moore, FR ' " Elements of Computer Music", Prentice-Hall '1990. If the acoustic model according to the singing voice signal analyzing unit 23 is a harmonic plus noise model, then one of the acoustic parameters corresponding to the sound box obtained after the analysis includes a set of harmonic parameters 6 1377557. Patent Specification No. 97148492 Correct the date of this revision: the number of September 27, 101 and a set of noise parameters; if the acoustic model according to the vocal signal analysis unit 23 is a phase speech coding model, the acoustic parameters corresponding to the above-mentioned sound frame are obtained after the analysis includes A set of frequency parameters and a set of intensity parameters. An acoustic parameter adjusting unit 24 adjusts the acoustic parameters according to the musical score information stored in the song database unit 21 to adjust the singing voice sung by the singer 10, for example, adjusting the parameters related to the pitch in the acoustic parameters to correct the pitch (pitch) Inaccurate problem, adjusting the parameters of the acoustic parameters in the above acoustic parameters to correct the problem of incorrect tempo, or even dithering the harmonic fundamental frequency corresponding to the acoustic parameters above a frequency value to add vibrato (vibrato) characteristics, finally get the complex adjusted acoustic parameters; it is worth noting that in order to keep the timbre of the singer (timbre) unchanged, in the above parameter adjustment process, the above acoustic parameters must be adjusted with the above The shape of the spectral envelope curve corresponding to the acoustic parameters remains the same. Next, if the acoustic signal is analyzed using a harmonic plus noise model, the song signal synthesis unit 25 further synthesizes the adjusted acoustic parameters into a plurality of synthesized syllables by a harmonic plus noise synthesis method; If the acoustic signal is analyzed using a phase speech coding model, the speech signal synthesizing unit 25 further synthesizes the adjusted acoustic parameters into a plurality of synthesized syllables by a phase speech coding synthesis method; the harmonic addition noise The synthesis method and the above-described phase speech coding synthesis method have been described in detail in the reference [1] '[2], so there is not much to be described here. Finally, the syllable of the above synthesized - is connected by a vocal signal concatenation unit 26 into a modified vocal signal. Fig. 2 is a flow chart showing a singing voice correcting method according to an embodiment of the present invention. At the beginning of the method, I received a singer from a singer who wants to sing 7 〇 / makeup _-year 9 ribs ton), so I got the song from the song library order, the second song, S2 〇 'the above music information Including the sound of each note ^ mouth. Then a syllable, a syllable, ==/ (steps the above-mentioned respective acoustic-acoustic models to t[23) and then the acoustic sounds of the above-mentioned respective sound frames; the reference sound = ^^ murmur model or phase The speech coding model, while the = order can be (4) references [1], [2]. If the root of the step S24 is categorized as a spectral add murmur model, then the acoustic _=== group murmur based on the acoustical number of the upper melody frame is calculated. Parameters; if the opposite is obtained from the opposite bank: 曰 coding model, then analyze and a set of sound intensity; the acoustic parameters of the X-day frame include - group frequency parameters. Next, according to the above-mentioned score information adjustment, the adjusted acoustic parameters (steps) S25), adjustment = to = the parameters related to the pitch in the acoustic parameters; tune: the parameters related to the length of the second parameter are corrected to the basic frequency corresponding to ±7 jitter of the beat value of the beat ί - let the singer Tone ": Sex; it is worth noting that for the acoustic parameters and the above-mentioned Bu, the acoustic parameters of the above-mentioned curve-shaped (four) axis must be the same as the m-sense/number of the spectral envelopes and the above acoustic model: after the adjustment according to the above (step gamma The most ^ = type to produce a plurality of synthesized syllables of the vocal signal (step =;) in series with the above synthesized syllable to obtain - correction 1377557 Patent No. 97148492 revised this revision period · H) 1 year 9 The third picture of the month 27曰 shows according to this A flowchart of a singing voice correction method according to another embodiment of the present invention. The method first receives song information from a singer who wants to sing, and then obtains music score information of the song from a song database (step S31). The above musical score information includes pitch, length and volume of each note. Then, a singing voice signal of the entire song sung by the singer is recorded (step S32), and the singing voice signal is divided into a plurality of syllables (step S33), and then the above Each syllable is divided into a sequence of sound boxes including a plurality of sound boxes (step S34), and then a predetermined acoustic analysis program is performed on each of the sound frames by an acoustic model to obtain a plurality of acoustic parameters corresponding to the acoustic model (step S35). The above acoustic model may be a harmonic plus noise model or a phase speech coding model, and the above-mentioned established acoustic analysis program can be referred to references [1], [2]. If the acoustic model based on step S35 is harmonic plus noise a model, the acoustic parameters corresponding to the sound box obtained after the analysis include a set of harmonic parameters and a set of noise parameters; if the acoustics according to step S35 The model is a phase speech coding model, and the acoustic parameters corresponding to the above sound frame are analyzed to include a set of frequency parameters and a set of sound intensity parameters. Next, the acoustic parameters are adjusted according to the above musical score information to obtain a plurality of adjusted acoustics. Parameter (step S36), the adjustment method is as follows: adjusting the parameters of the pitch in the acoustic parameters to correct the inaccuracy of the pitch, adjusting the parameters of the acoustic parameters in the acoustic parameters to correct the problem of the beat, or even The basic frequency corresponding to the acoustic parameter is shaken above and below the frequency value to add the characteristics of the vibrato; it is worth noting that in order to keep the timbre of the singer unchanged, in step S36, the acoustic parameters must be adjusted with the above. The acoustic envelope parameters corresponding to each other have the same spectral envelope curve shape. After the parameter adjustment is completed, a plurality of synthesized syllables are generated according to the above-mentioned adjusted acoustic parameters and the above acoustic model (step 9 1377557 • Patent Specification No. 97148492 Revision Date: September 27, Sept. S37). Finally, the above synthesized syllable is concatenated to obtain a corrected vocal signal (step S38). The present invention has been described above by way of example, and is not intended to limit the scope of the present invention, and the invention may be modified and modified without departing from the spirit and scope of the invention. The scope of protection is subject to the definition of the scope of the patent application attached. 10 1377557 Amendment date: September 27, 101. Patent Specification Revision No. 97148492 [Simplified Description of the Drawings] Fig. 1 shows a singing voice correcting device according to an embodiment of the present invention. Fig. 2 is a flow chart showing a singing voice correcting method according to an embodiment of the present invention. Fig. 3 is a flow chart showing a singing voice correcting method according to another embodiment of the present invention. [Main component symbol description] 10~Singer; 20~Song correction device; 21~ song database unit; 22~ pronunciation recording unit; 23~ singing signal analysis unit; 24~ acoustic parameter adjustment unit; 25~ singing signal synthesis unit ; 26 ~ singing signal concatenation unit.