TWI579716B - Two - level phrase search system and method - Google Patents
Two - level phrase search system and method Download PDFInfo
- Publication number
- TWI579716B TWI579716B TW104140064A TW104140064A TWI579716B TW I579716 B TWI579716 B TW I579716B TW 104140064 A TW104140064 A TW 104140064A TW 104140064 A TW104140064 A TW 104140064A TW I579716 B TWI579716 B TW I579716B
- Authority
- TW
- Taiwan
- Prior art keywords
- phrase
- level
- correlation coefficient
- module
- candidate
- Prior art date
Links
Description
本發明係關於一種二層級樂句搜尋系統及方法,特別是關於一種能夠在快速比對機制和精準比對機制間取得平衡之二層級樂句搜尋系統及方法。 The present invention relates to a two-level phrase search system and method, and more particularly to a two-level phrase search system and method capable of balancing a fast comparison mechanism and a precision comparison mechanism.
目前,隨著音樂數位化的普及,音樂數據正以驚人的速度不斷增長。其中,樂器數位介面檔案(Musical Instrument Digital Interface,MIDI)被廣泛使用以簡化音樂的表達,造就許多巨量的樂器數位介面檔案資料庫,該如何從中進行搜尋則成為熱門的技術。 At present, with the popularity of digital music, music data is growing at an alarming rate. Among them, the Musical Instrument Digital Interface (MIDI) is widely used to simplify the expression of music, and has created a huge number of instrument digital interface file database. How to search from it becomes a popular technology.
傳統利用文字檢索搜尋的方式,僅能針對音樂的額外標註進行搜尋,並不能滿足使用者想要直接透過旋律搜尋音樂的需求,因此,許多以樂句搜尋的方法被發明以改善這個問題。 Traditionally, the use of text search and search can only search for additional annotations of music, and it does not satisfy the user's need to search for music directly through melody. Therefore, many methods of searching for phrases are invented to improve this problem.
由於音樂樂句的產生方法並非唯一,加上搜尋樂句和資料庫樂句的來源可能有異,在做樂句比對的時候,需要額外的運算才能更正確地衡量兩樂句間的相似程度。各式的音樂樂句比對機制被發明,如字符串相似度(Levenshtein distance)、最長公共子序列(LCS,longest common subsequence)、線性伸縮(linear scaling)、動態時間校正(DTW,dynamic time warping)等,如何從中選出恰當的比對機制常意味著運算複雜度和準確度間的取捨。 Since the production method of musical phrases is not unique, and the sources of search phrases and database phrases may be different, when doing phrase comparison, additional calculations are needed to more accurately measure the similarity between the two phrases. Various musical phrase comparison mechanisms were invented, such as string similarity (Levenshtein distance), longest common subsequence (LCS, longest common Subsequence), linear scaling, dynamic time warping (DTW), etc. How to choose the appropriate comparison mechanism often means the trade-off between computational complexity and accuracy.
較簡易的方法如線性伸縮能帶來較快的反應速度,但也伴隨 著較粗糙的搜尋結果,而較複雜的方法如動態時間校正則以較長的時間換得較精確的搜尋結果。 Simpler methods such as linear stretching can result in faster response times, but also Rough search results, while more complex methods such as dynamic time correction are used to exchange more accurate search results for a longer period of time.
有鑑於上述習知技藝之問題,本發明之目的就是在提供一種二層級樂句搜尋系統及方法,以解決並改善當前搜尋樂句時難以於搜尋速率及精準度之間進行取捨之議題。 In view of the above-mentioned problems of the prior art, the object of the present invention is to provide a two-level phrase search system and method for solving and improving the problem of difficulty in selecting between search rate and accuracy when searching for a phrase.
本發明之樂句搜尋系統包含一樂句擷取模組、一樂句關聯建立模組、一樂句資料庫、一第一層級樂句比對模組以及一第二層級樂句比對模組。樂句擷取模組自符號式音樂檔案(MIDI)擷取樂句片段;樂句關聯建立模組計算樂句片段之樂句關聯係數;樂句資料庫儲存待比對樂句片段之資訊及樂句關聯係數;第一層級樂句比對模組從樂句資料庫篩選出第一層級候選樂句集合,並比對出與待比對樂句片段最相關之樂句集合;第二層級樂句比對模組根據第一層級樂句比對模組之比對結果以及樂句資料庫之樂句關聯係數,篩選出第二層級候選樂句集合,並自第二層級候選樂句集合比對出與待比對樂句片段最相關之樂句。 The phrase search system of the present invention comprises a phrase extraction module, a phrase association building module, a phrase database, a first level phrase comparison module and a second level phrase comparison module. The phrase capture module extracts the phrase segment from the symbolic music file (MIDI); the phrase association module calculates the phrase correlation coefficient of the phrase segment; the phrase database stores the information of the phrase segment to be compared and the phrase correlation coefficient; the first level The phrase comparison module filters out the first-level candidate phrase set from the phrase database, and compares the phrase set most relevant to the to-be-matched phrase segment; the second-level phrase comparison module is based on the first-level phrase comparison model. The comparison result of the group and the phrase correlation coefficient of the phrase database select a second level candidate phrase set, and compare the phrase most relevant to the to-be-matched phrase segment from the second level candidate phrase set.
本發明之樂句搜尋方法包含下列步驟:以樂句擷取模組自符號式音樂檔案(MIDI)擷取樂句片段;以樂句關聯建立模組計算樂句片段之樂句關聯係數;以樂句資料庫儲存待比對樂句片段之資訊及樂句關聯係數;以第一層級樂句比對模組從樂句資料庫篩選出第一層級候選樂句集合,並比對出與待比對樂句片段最相關之樂句集合;以及以第二層級樂句比對模組根據第一層級樂句比對模組之比對結果以及樂句資料庫之樂句關聯係數,篩選出第二層級候選樂句集合,並自第二層級候選樂句集合比對出與待比對樂句片段最相關之樂句。 The phrase search method of the present invention comprises the following steps: extracting a phrase segment from a symbolic music file (MIDI) by using a phrase capture module; and calculating a phrase correlation coefficient of the phrase segment by using a phrase association module; storing the waiting ratio in the phrase database Information about the phrase segment and the phrase correlation coefficient; the first level phrase matching module selects the first level candidate phrase set from the phrase database, and compares the phrase set most relevant to the to-be-matched phrase segment; The second level phrase comparison module selects the second level candidate phrase set according to the comparison result of the first level phrase comparison module and the phrase correlation coefficient of the phrase database, and compares the second level candidate phrase set from the second level candidate phrase set. The phrase most relevant to the segment of the phrase to be compared.
承上所述,依本發明之二層級樂句搜尋系統及方法,其可具有一或多個下述優點: In view of the above, a two-level phrase search system and method in accordance with the present invention may have one or more of the following advantages:
1.關聯係數建立:預先分析音樂資料庫,並建立音樂樂句間之關聯係數,與快速搜尋之搜尋結果結合後,可有效限縮精準搜尋之搜尋範圍,達到兼俱速度及精準度之功效。 1. Correlation coefficient establishment: Pre-analyze the music database and establish the correlation coefficient between music phrases. When combined with the search results of fast search, it can effectively limit the search range of accurate search, and achieve the effect of speed and precision.
2.快速搜尋:於第一層級提供快捷之音樂樂句搜尋方法,得以於短暫的時間內提供簡略的搜尋結果給使用者。 2. Quick Search: Provides a quick music phrase search method at the first level, which provides a short search result to the user in a short period of time.
3.精準搜尋:於第二層級提供精準之音樂樂句搜尋方法,利用快速搜尋之結果與預先算好的關聯係數,找出準確之搜尋結果給使用者。 3. Accurate search: Provide accurate music phrase search methods at the second level, and use the results of quick search and pre-calculated correlation coefficients to find accurate search results for users.
10‧‧‧樂句搜尋系統 10‧‧‧ Phrase Search System
100‧‧‧樂句擷取模組 100‧‧‧ phrase extraction module
101‧‧‧輸入音訊訊號 101‧‧‧ Input audio signal
110‧‧‧查詢樂句片段數字序列 110‧‧‧Query phrase sequence number sequence
200‧‧‧樂句關聯建立模組 200‧‧‧ phrase association module
210‧‧‧樂句資料庫存取模組 210‧‧‧ Phrase data inventory module
220‧‧‧樂句關聯計算模組 220‧‧‧ Phrase Correlation Computation Module
221‧‧‧樂句關聯係數演算法 221‧‧‧ phrase correlation coefficient algorithm
300‧‧‧樂句資料庫 300‧‧‧ Phrase Database
310‧‧‧樂句片段資訊資料表 310‧‧‧ Phrase Clip Information Sheet
311‧‧‧音樂編號 311‧‧‧ music number
312‧‧‧片段編號 312‧‧‧Segment number
313‧‧‧樂句片段數字序列 313‧‧‧ Phrase segment number sequence
320‧‧‧樂句關聯係數資料表 320‧‧‧ Phrase Correlation Coefficient Data Sheet
321‧‧‧來源音樂編號 321‧‧‧Source music number
322‧‧‧來源片段編號 322‧‧‧Source segment number
323‧‧‧目的音樂編號 323‧‧‧Object music number
324‧‧‧目的片段編號 324‧‧‧ Target segment number
325‧‧‧關聯係數 325‧‧‧Correlation coefficient
400‧‧‧第一層級樂句比對模組 400‧‧‧First level phrase comparison module
410‧‧‧第一層級候選樂句集合篩選模組 410‧‧‧First-level candidate phrase collection screening module
411‧‧‧第一層級候選樂句集合篩選演算法 411‧‧‧First-level candidate phrase collection screening algorithm
412‧‧‧第一層級候選樂句集合 412‧‧‧First Level Candidate Phrase Collection
420‧‧‧第一層級樂句關聯係數計算模組 420‧‧‧First Level Phrase Correlation Coefficient Calculation Module
421‧‧‧第一層級樂句關聯係數演算法 421‧‧‧The first level of phrase correlation coefficient algorithm
422‧‧‧第一層級候選樂句關聯係數集合 422‧‧‧First-level candidate phrase correlation coefficient set
430‧‧‧第一層級樂句查詢結果篩選模組 430‧‧‧First level phrase query result screening module
431‧‧‧第一層級樂句查詢結果篩選演算法 431‧‧‧The first level of phrase query result screening algorithm
432‧‧‧第一層級候選樂句查詢結果 432‧‧‧First level candidate phrase query results
500‧‧‧第二層級樂句比對模組 500‧‧‧Second level phrase comparison module
510‧‧‧查詢擴張模組 510‧‧‧Query expansion module
511‧‧‧第二層級候選樂句集合 511‧‧‧Second Level Candidate Phrase Collection
520‧‧‧第二層級樂句關聯係數計算模組 520‧‧‧Second level phrase correlation coefficient calculation module
521‧‧‧第二層級樂句關聯係數演算法 521‧‧‧Second level phrase correlation coefficient algorithm
522‧‧‧第二層級候選樂句關聯係數集合 522‧‧‧Second Level Candidate Phrase Correlation Coefficient Set
530‧‧‧第二層級樂句查詢結果篩選模組 530‧‧‧Second level phrase query result screening module
531‧‧‧第二層級樂句查詢結果篩選演算法 531‧‧‧Second level phrase query result screening algorithm
532‧‧‧第二層級候選樂句查詢結果 532‧‧‧Second Level Candidate Phrase Search Results
圖1為本發明之樂句搜尋系統之系統架構圖。 1 is a system architecture diagram of a phrase search system of the present invention.
圖2為本發明之樂句搜尋系統之樂句擷取模組及其輸入示意圖。 2 is a schematic diagram of a phrase capture module of the phrase search system of the present invention and an input thereof.
圖3為本發明之樂句搜尋系統之樂句資料庫示意圖。 3 is a schematic diagram of a phrase database of the phrase search system of the present invention.
圖4為本發明之樂句搜尋系統之樂句關聯建立模組及樂句資料庫關聯圖。 4 is a diagram showing a phrase association building module and a phrase database association of the phrase search system of the present invention.
圖5為本發明之樂句搜尋系統之第一層級樂句比對模組與其它資料關聯圖。 FIG. 5 is a diagram showing the relationship between the first level phrase comparison module and other data of the phrase search system of the present invention.
圖6為本發明之樂句搜尋系統之第二層級樂句比對模組與其它資料關聯圖。 FIG. 6 is a diagram showing the second level phrase comparison module and other data association diagrams of the phrase search system of the present invention.
本發明之樂句搜尋系統10主要元件如圖1及圖2所示,其係由五個系統元件所組成,其名稱與主要功能分述如下:樂句擷取模組100自樂器數位介面檔案(MIDI)擷取樂句片段;樂句關聯建立模組200計算兩兩樂句間之關聯係數;樂句資料庫300儲存欲比對之樂句片段資訊(又或稱待比對樂句片段資訊)及樂句關聯係數;第一層級樂句比對模組400自樂句資料庫篩選出第一層級候選樂句集合,並比對出與查詢樂句最相關之樂句集合;第二層級樂句比對模組500根據第一層級樂句比對模組之結果,以及樂句資料庫之樂句關聯係數,篩選出第二層級候選樂句集合。並自第二層級候選樂句集合比對出與查詢樂句最相關之樂句。 The main components of the phrase search system 10 of the present invention are as shown in FIG. 1 and FIG. 2, which are composed of five system components. The names and main functions are described as follows: the phrase capture module 100 is from the instrument digital interface file (MIDI). Taking a phrase segment; the phrase association building module 200 calculates a correlation coefficient between the two phrases; the phrase database 300 stores the information of the phrase segment to be compared (or the information of the phrase segment) and the phrase correlation coefficient; The first-level phrase comparison module 400 filters out the first-level candidate phrase set from the phrase database, and compares the phrase set most relevant to the query phrase; the second-level phrase comparison module 500 compares the first-level phrase according to the first-level phrase. The result of the module, and the phrase correlation coefficient of the phrase database, selects the second level candidate phrase set. And the phrase most relevant to the query phrase is compared from the second level candidate phrase set.
請參考圖3,本發明的樂句資料庫300包含了兩組主要資料表,其一是樂句片段資訊資料表310,其二是樂句關聯係數資料表320。樂句片段資訊資料表310之資料欄位包含音樂編號311、片段編號312以及樂句片段數字序列313。樂句關聯係數資料表320之資料欄位包含了來源音樂 編號321、來源片段編號322、目的音樂編號323、目的片段編號324以及關聯係數325。 Referring to FIG. 3, the phrase database 300 of the present invention includes two sets of main data tables, one of which is a phrase segment information data table 310, and the other is a phrase correlation coefficient data table 320. The data field of the phrase segment information table 310 includes a music number 311, a segment number 312, and a phrase segment number sequence 313. The data field of the phrase correlation coefficient data table 320 contains the source music. No. 321, source segment number 322, destination music number 323, destination segment number 324, and correlation coefficient 325.
樂句關聯建立模組200包含了樂句資料庫存取模組210、以 及樂句關聯計算模組220。樂句資料庫存取模組210負責連接及存取修改樂句資料庫300中之樂句片段資訊資料表310以及樂句關聯係數資料表320。 樂句關聯計算模組220以樂句資料庫存取模組210提取任意兩樂句資訊,以樂句關聯係數演算法221計算樂句間之關聯係數,並將來源音樂編號、來源樂句編號、目的音樂編號、目的樂句編號以及關聯係數儲存至樂句關聯係數資料表320。 The phrase association building module 200 includes a phrase material inventory taking module 210, And the phrase correlation calculation module 220. The phrase data inventory retrieval module 210 is responsible for connecting and accessing the phrase segment information data table 310 and the phrase correlation coefficient data table 320 in the modified phrase database 300. The phrase association calculation module 220 extracts any two phrase information by the phrase data inventory module 210, and calculates the correlation coefficient between the phrases by the phrase correlation coefficient algorithm 221, and the source music number, the source phrase number, the destination music number, and the destination phrase. The number and associated coefficient are stored in the phrase correlation coefficient data table 320.
樂句擷取模組100將輸入之音訊訊號101或符號式音高時間 序列102轉換為樂句資料庫中樂句片段數字序列儲存之格式,成為查詢樂句片段數字序列110,以進行接續之比對搜尋。 The phrase capture module 100 will input the audio signal 101 or the symbolic pitch time. The sequence 102 is converted into a format in which the sequence number sequence of the phrase segments in the phrase database is stored, and becomes a query phrase segment number sequence 110 for subsequent comparison search.
請參考圖5,第一層級樂句比對模組400包含了第一層級候 選樂句集合篩選模組410、第一層級樂句關聯係數計算模組420以及第一層級樂句查詢結果篩選模組430。第一層級候選樂句集合篩選模組410根據第一層級候選樂句集合篩選演算法411,自樂句片段資訊資料表310篩選出第一層級候選樂句集合412。第一層級樂句關聯係數計算模組420依序以第一層級樂句關聯係數演算法421計算查詢樂句片段數字序列110以及第一層級候選樂句集合412之第一層級候選樂句關聯係數集合422。第一層級樂句查詢結果篩選模組430依據第一層級候選樂句關聯係數集合422,以第一層級樂句查詢結果篩選演算法431,篩選出第一層級候選樂句查詢結果432。 Referring to FIG. 5, the first level phrase comparison module 400 includes the first level candidate. The selected phrase collection screening module 410, the first level phrase correlation coefficient calculation module 420, and the first level phrase query result screening module 430. The first level candidate phrase set screening module 410 filters the algorithm 411 according to the first level candidate phrase set, and selects the first level candidate phrase set 412 from the phrase segment information table 310. The first level phrase correlation coefficient calculation module 420 sequentially calculates the query phrase segment number sequence 110 and the first level candidate phrase correlation coefficient set 422 of the first level candidate phrase set 412 by the first level phrase correlation coefficient algorithm 421. The first level phrase query result screening module 430 selects the first level candidate phrase query result 432 according to the first level candidate phrase correlation coefficient set 422, and selects the first level candidate phrase query result 432.
請參考圖6,第二層級樂句比對模組500包含了查詢擴張模 組510、第二層級樂句關聯係數計算模組520以及第二層級樂句查詢結果篩選模組530。查詢擴張模組510以第一層級候選樂句查詢結果432依序查詢樂句關聯係數資料表320,篩選出第二層級候選樂句集合511。第二層級樂句關聯係數計算模組520依序以第二層級樂句關聯係數演算法521計算查詢樂句片段數字序列110以及第二層級候選樂句集合511之第二層級候選樂句關聯係數集合522。第二層級樂句查詢結果篩選模組530依據第二層級候選樂句關聯係數集合522,以第二層級樂句查詢結果篩選演算法531,篩選出第二層級候選樂句查詢結果532,作為此次查詢之結果。 Referring to FIG. 6, the second level phrase comparison module 500 includes a query expansion module. The group 510, the second level phrase correlation coefficient calculation module 520 and the second level phrase query result screening module 530. The query expansion module 510 sequentially queries the phrase correlation coefficient data table 320 with the first level candidate phrase query result 432 to filter out the second level candidate phrase set 511. The second level phrase correlation coefficient calculation module 520 sequentially calculates the second layer candidate phrase correlation coefficient set 522 of the query phrase segment number sequence 110 and the second level candidate phrase set 511 by the second level phrase correlation coefficient algorithm 521. The second level phrase query result screening module 530 selects the second level candidate phrase query result 531 according to the second level candidate phrase correlation coefficient set 522, and selects the second level candidate phrase query result 532 as the result of the query. .
以下將以各圖式來說明本揭露之樂句搜尋系統的運作機制。 The operation mechanism of the phrase search system of the present disclosure will be described below in the drawings.
步驟一:如圖4,樂句關聯計算模組220以樂句資料庫存取模組210提取任意兩樂句資訊,以樂句關聯係數演算法221計算樂句間之關聯係數,並將來源音樂編號、來源樂句編號、目的音樂編號、目的樂句編號以及關聯係數儲存至樂句關聯係數資料表320。其中樂句關聯係數演算法221可以使用平均數/中位數將兩樂句進行正規化,並使用動態時間校正的音高輪廓比對方法求得兩樂句之距離,取倒數後成為關聯係數。例如樂句A之數字序列為「10、10、10、11、11」,樂句B之數字序列為「19、20、20、20、22、21」,先將樂句B正規化為樂句B’「9、10、10、10、12、11」,再透過動態時間校正計算出樂句A與B’之差距為一個刪除運算以及一個取代運算,距離為2,關聯係數即為0.5。 Step 1: As shown in FIG. 4, the phrase correlation calculation module 220 extracts any two phrase information by the phrase data inventory taking module 210, calculates the correlation coefficient between the phrases by the phrase correlation coefficient algorithm 221, and numbers the source music number and the source phrase number. The destination music number, the destination phrase number, and the correlation coefficient are stored in the phrase correlation coefficient data table 320. The phrase correlation coefficient algorithm 221 can normalize the two phrases using the mean/median, and obtain the distance between the two phrases using the dynamic time corrected pitch contour comparison method, and obtain the correlation coefficient after taking the reciprocal. For example, the sequence of numbers in phrase A is "10, 10, 10, 11, 11", and the sequence of numbers in phrase B is "19, 20, 20, 20, 22, 21". The phrase B is first normalized to phrase B'" 9, 10, 10, 10, 12, 11", and then through the dynamic time correction to calculate the difference between the phrase A and B' is a delete operation and a substitution operation, the distance is 2, the correlation coefficient is 0.5.
步驟二:樂句擷取模組100將輸入音訊訊號或符號式音高時間序列轉換為樂句資料庫中樂句片段數字序列儲存之格式,成為查詢樂句片段數字序列110,以進行接續之比對搜尋。查詢樂句片段數字序列110為 固定時間間隔取樣得到的主要聲音頻率值,例如連續一秒的中央Do經間隔200ms取樣而成的數字序列為「261.6、261.6、261.6、261.6、261.6」。 Step 2: The phrase capture module 100 converts the input audio signal or symbolic pitch time series into a format of the sequence of the phrase segment digital sequence stored in the phrase database, and becomes a query phrase segment digital sequence 110 for subsequent comparison search. Query phrase segment number sequence 110 is The main sound frequency values obtained by sampling at fixed time intervals, for example, the serial sequence of the central Do that is sampled at intervals of 200 ms for one second is "261.6, 261.6, 261.6, 261.6, 261.6".
步驟三:第一層級候選樂句集合篩選模組410根據第一層級候選樂句集合篩選演算法411,自樂句片段資訊資料表310篩選出第一層級候選樂句集合412。其中第一層級候選樂句集合篩選演算法411可以隨機從樂句片段資訊資料表310中取出一樂句片段子集合而成;第一層級候選樂句集合篩選演算法411亦可根據搜尋條件自樂句片段資訊資料表310中取得特定音樂編號的子集合。 Step 3: The first level candidate phrase set screening module 410 filters the algorithm 411 according to the first level candidate phrase set, and selects the first level candidate phrase set 412 from the phrase segment information table 310. The first level candidate phrase set screening algorithm 411 may randomly take out a phrase segment subset from the phrase segment information table 310; the first level candidate phrase set screening algorithm 411 may also use the phrase segment information according to the search condition. A subset of the specific music numbers is obtained in table 310.
步驟四:第一層級樂句關聯係數計算模組420依序以第一層級關聯係數演算法421計算查詢樂句片段數字序列110以及第一層級候選樂句集合412之第一層級候選樂句關聯係數集合422。其中第一層級樂句關聯係數演算法421可以使用平均數/中位數將查詢樂句片段數字序列110以及第一層級候選樂句進行正規化,並計算兩者間的歐氏距離,取倒數後成為第一層級候選樂句關聯係數。 Step 4: The first level phrase correlation coefficient calculation module 420 sequentially calculates the query phrase segment number sequence 110 and the first level candidate phrase correlation coefficient set 422 of the first level candidate phrase set 412 by the first level correlation coefficient algorithm 421. The first level phrase correlation coefficient algorithm 421 can normalize the query phrase segment number sequence 110 and the first level candidate phrase using the mean/median, and calculate the Euclidean distance between the two, and take the countdown to become the first One-level candidate phrase correlation coefficient.
步驟五:第一層級樂句查詢結果篩選模組430依據第一層級候選樂句關聯係數集合422,以第一層級樂句查詢結果篩選演算法431,篩選出第一層級候選樂句查詢結果432。其中第一層級樂句查詢結果篩選演算法431可依據關聯係數由大到小排序,取得前十名作為查詢結果。 Step 5: The first level phrase query result screening module 430 selects the first level candidate phrase query result 431 according to the first level candidate phrase correlation coefficient set 422, and selects the first level candidate phrase query result 432. The first level phrase query result screening algorithm 431 can be sorted according to the correlation coefficient from large to small, and the top ten is obtained as the query result.
步驟六:查詢擴張模組510以第一層級候選樂句查詢結果432依序查詢樂句關聯係數資料表320,篩選出第二層級候選樂句集合512。其中第二層級候選樂句集合511篩選方式可為所有第一層級候選樂句查詢結果432在樂句關聯係數資料表320關聯係數最高之前一百名之樂句集合。 Step 6: The query expansion module 510 sequentially queries the phrase correlation coefficient data table 320 with the first level candidate phrase query result 432, and filters out the second level candidate phrase set 512. The second level candidate phrase set 511 screening mode may be a set of one hundred words before all the first level candidate phrase query result 432 has the highest correlation coefficient in the phrase correlation coefficient data table 320.
步驟七:第二層級樂句關聯係數計算模組520依序以第二層級樂句關聯係數演算法521計算查詢樂句片段數字序列110以及第二層級候選樂句集合511之第二層級候選樂句關聯係數集合522。其中第二層級關聯係數演算法521可以與樂句關聯係數演算法221相同。 Step 7: The second level phrase correlation coefficient calculation module 520 sequentially calculates the query phrase segment number sequence 110 and the second level candidate phrase correlation coefficient set 522 of the second level candidate phrase set 511 by the second level phrase correlation coefficient algorithm 521. . The second level correlation coefficient algorithm 521 may be the same as the phrase correlation coefficient algorithm 221.
步驟八:第二層級樂句查詢結果篩選模組530依據第二層級候選樂句關聯係數集合522,以第二層級樂句查詢結果篩選演算法531,篩選出第二層級候選樂句查詢結果532,作為此次查詢之結果。其中第二層級樂句查詢結果篩選演算法532可依據關聯係數由大到小排序,取得前十名作為查詢結果。 Step 8: The second level phrase query result screening module 530 selects the second level candidate phrase query result 531 according to the second level candidate phrase correlation coefficient set 522, and selects the second level candidate phrase query result 532 as the current The result of the query. The second level phrase query result screening algorithm 532 can be sorted according to the correlation coefficient from large to small, and the top ten is obtained as the query result.
綜上所述,本發明之二層級樂句搜尋系統及方法,具有創新性、立即性、效率性及可擴充性等優點,不但在技術思想上確屬創新,並能有效快速且精確地查詢音樂樂句。 In summary, the two-level phrase search system and method of the present invention have the advantages of innovation, immediateness, efficiency, and expandability, and are not only innovative in terms of technical ideas, but also can efficiently and quickly and accurately query music. Phrases.
以上所述僅為舉例性,而非為限制性者。任何未脫離本發明之精神與範疇,而對其進行之等效修改或變更,均應包含於後附之申請專利範圍中。 The above is intended to be illustrative only and not limiting. Any equivalent modifications or alterations to the spirit and scope of the invention are intended to be included in the scope of the appended claims.
10‧‧‧樂句搜尋系統 10‧‧‧ Phrase Search System
100‧‧‧樂句擷取模組 100‧‧‧ phrase extraction module
200‧‧‧樂句關聯建立模組 200‧‧‧ phrase association module
300‧‧‧樂句資料庫 300‧‧‧ Phrase Database
400‧‧‧第一層級樂句比對模組 400‧‧‧First level phrase comparison module
500‧‧‧第二層級樂句比對模組 500‧‧‧Second level phrase comparison module
Claims (10)
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
TW104140064A TWI579716B (en) | 2015-12-01 | 2015-12-01 | Two - level phrase search system and method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
TW104140064A TWI579716B (en) | 2015-12-01 | 2015-12-01 | Two - level phrase search system and method |
Publications (2)
Publication Number | Publication Date |
---|---|
TWI579716B true TWI579716B (en) | 2017-04-21 |
TW201721464A TW201721464A (en) | 2017-06-16 |
Family
ID=59240921
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
TW104140064A TWI579716B (en) | 2015-12-01 | 2015-12-01 | Two - level phrase search system and method |
Country Status (1)
Country | Link |
---|---|
TW (1) | TWI579716B (en) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP1307833B1 (en) * | 2000-07-31 | 2006-06-07 | Landmark Digital Services LLC | Method for search in an audio database |
TW200745896A (en) * | 2006-06-07 | 2007-12-16 | Cweb Technology Inc | Method for retrieving song by tempo signal |
TW200813759A (en) * | 2006-05-23 | 2008-03-16 | Creative Tech Ltd | A method and apparatus for accessing an audio file from a collection of audio files using tonal matching |
US20090119614A1 (en) * | 2007-11-07 | 2009-05-07 | Nokia Corporation | Method, Apparatus and Computer Program Product for Heirarchical Navigation with Respect to Content Items of a Media Collection |
CN101520808A (en) * | 2008-02-29 | 2009-09-02 | 索尼株式会社 | Method for visualizing audio data |
TW201101061A (en) * | 2009-06-19 | 2011-01-01 | Ipeer Multimedia Internat Ltd | Multimedia identification system and method, and the application |
-
2015
- 2015-12-01 TW TW104140064A patent/TWI579716B/en active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP1307833B1 (en) * | 2000-07-31 | 2006-06-07 | Landmark Digital Services LLC | Method for search in an audio database |
TW200813759A (en) * | 2006-05-23 | 2008-03-16 | Creative Tech Ltd | A method and apparatus for accessing an audio file from a collection of audio files using tonal matching |
TW200745896A (en) * | 2006-06-07 | 2007-12-16 | Cweb Technology Inc | Method for retrieving song by tempo signal |
US20090119614A1 (en) * | 2007-11-07 | 2009-05-07 | Nokia Corporation | Method, Apparatus and Computer Program Product for Heirarchical Navigation with Respect to Content Items of a Media Collection |
CN101520808A (en) * | 2008-02-29 | 2009-09-02 | 索尼株式会社 | Method for visualizing audio data |
TW201101061A (en) * | 2009-06-19 | 2011-01-01 | Ipeer Multimedia Internat Ltd | Multimedia identification system and method, and the application |
Also Published As
Publication number | Publication date |
---|---|
TW201721464A (en) | 2017-06-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Bertin-Mahieux et al. | Large-scale cover song recognition using hashed chroma landmarks | |
US8489606B2 (en) | Music search apparatus and method using emotion model | |
WO2019096068A1 (en) | Voice recognition and error correction method and voice recognition and error correction system | |
Mantena et al. | Speed improvements to information retrieval-based dynamic time warping using hierarchical k-means clustering | |
Guo et al. | Time-Warped Longest Common Subsequence Algorithm for Music Retrieval. | |
Guo et al. | A query by humming system based on locality sensitive hashing indexes | |
Wang et al. | An effective and efficient method for query by humming system based on multi-similarity measurement fusion | |
KR101593961B1 (en) | Apparatus and method for music search using emotion model | |
Putri et al. | Music information retrieval using Query-by-humming based on the dynamic time warping | |
Yao et al. | An efficient cascaded filtering retrieval method for big audio data | |
Zhang et al. | Low-latency online speaker diarization with graph-based label generation | |
TWI579716B (en) | Two - level phrase search system and method | |
Dumpala et al. | Analysis of constraints on segmental DTW for the task of query-by-example spoken term detection | |
Nam et al. | A new query‐by‐humming system based on the score level fusion of two classifiers | |
Vaglio et al. | The words remain the same: Cover detection with lyrics transcription | |
US10776420B2 (en) | Fingerprint clustering for content-based audio recognition | |
Rajan et al. | Oktoechos classification in liturgical music using musical texture features | |
Xi et al. | The design and construction based on the ASEAN piano music library and display platform | |
Lin et al. | Advanced query by humming system using diffused hidden Markov model and tempo based dynamic programming | |
Mostafa et al. | A machine learning based music retrieval and recommendation system | |
Deepsheka et al. | Recurrent neural network based music recognition using audio fingerprinting | |
Vasudevan et al. | A Hybrid Cluster-Classifier model for Carnatic Raga Classification | |
Kharat et al. | A survey on query by singing/humming | |
Weck et al. | WikiMuTe: A web-sourced dataset of semantic descriptions for music audio | |
Zhou et al. | Research on matching method in humming retrieval |