Citation Context

...e of pi to all the vectors in the database. However, this results in a search time that depends linearly on the database size. To obtain a sublinear time complexity, we use locality sensitive hashing =-=[8, 9]-=-. LSH is a randomized algorithm for searching approximately nearest neighbors in high dimension spaces. The idea is that the points whose distances are within the threshold r will be hashed to a same ...

Citation Context

...e of pi to all the vectors in the database. However, this results in a search time that depends linearly on the database size. To obtain a sublinear time complexity, we use locality sensitive hashing =-=[8, 9]-=-. LSH is a randomized algorithm for searching approximately nearest neighbors in high dimension spaces. The idea is that the points whose distances are within the threshold r will be hashed to a same ...

Citation Context

...melodic similarity. There exist lots of QBH studies in the literature. Most approaches use either note or pitch sequence to represent the query. Matching approaches include string matching techniques =-=[2]-=-, hidden Markov models [3, 4], and dynamic programming [5]. A number of the current state-of-the-art QBH systems have been evaluated in Music Information Retrieval Evaluation eXchange (MIREX) 2006 and...

Citation Context

...ciated with automatic transcription of a query into temporally segmented note events or into frame-wise measured pitch trajectory, whereas the latter concentrates on measuring melodic similarity. See =-=[1]-=- for an overview of music information retrieval systems and research on melodic similarity. There exist lots of QBH studies in the literature. Most approaches use either note or pitch sequence to repr...

Citation Context

...it is not necessary to handle polyphony in query transcription, this method is used also to produce a melody database directly from music recordings (see Fig. 1). The method is an improved version of =-=[12]-=-. Briefly, the method uses a frame-wise pitch salience estimator to measure the strength of different fundamental frequencies in 92.9 ms analysis frames with 23.2 ms interval between successive frames...

Citation Context

..., we built a server for handling the melodic fragment index and a client for retrieving similar melodic fragments from the database. Recently, LSH has been applied, e.g., in remixed music recognition =-=[10]-=- and in audio fingerprinting [11]. 3. QUERY PROCESSING AND MELODY RETRIEVAL The retrieval stage consists of the following steps: i) transcription of a sung query into notes, ii) extraction of pitch ve...

Citation Context

... literature. Most approaches use either note or pitch sequence to represent the query. Matching approaches include string matching techniques [2], hidden Markov models [3, 4], and dynamic programming =-=[5]-=-. A number of the current state-of-the-art QBH systems have been evaluated in Music Information Retrieval Evaluation eXchange (MIREX) 2006 and 2007. The major challenges for QBH systems include i) han...

Citation Context

...l space where the distance is given by ‖pi − pj‖ = dX |pi(k) − pj(k)| 2 k=1 !(1/2) . (1) Euclidean distance is not only simple but also appears to be very effective measure for similarity (see, e.g., =-=[7]-=-). Given a melodic fragment defined by point pi, we can find similar fragments in the index by searching for nearest neighbors (NNs) of the point, i.e., all the points to which the distance is less th...

Citation Context

...exist lots of QBH studies in the literature. Most approaches use either note or pitch sequence to represent the query. Matching approaches include string matching techniques [2], hidden Markov models =-=[3, 4]-=-, and dynamic programming [5]. A number of the current state-of-the-art QBH systems have been evaluated in Music Information Retrieval Evaluation eXchange (MIREX) 2006 and 2007. The major challenges f...

Citation Context

...the melodic fragment index and a client for retrieving similar melodic fragments from the database. Recently, LSH has been applied, e.g., in remixed music recognition [10] and in audio fingerprinting =-=[11]-=-. 3. QUERY PROCESSING AND MELODY RETRIEVAL The retrieval stage consists of the following steps: i) transcription of a sung query into notes, ii) extraction of pitch vectors from the notes, iii) retrie...

Citation Context

...exist lots of QBH studies in the literature. Most approaches use either note or pitch sequence to represent the query. Matching approaches include string matching techniques [2], hidden Markov models =-=[3, 4]-=-, and dynamic programming [5]. A number of the current state-of-the-art QBH systems have been evaluated in Music Information Retrieval Evaluation eXchange (MIREX) 2006 and 2007. The major challenges f...

Citation Context

...atic melody transcription method to produce melody database directly from music recordings and achieve very encouraging results. Experimental results for QBH of audio were reported only very recently =-=[6]-=-. 1-4244-1484-9/08/$25.00 ©2008 IEEE 2249 ICASSP 2008MIDI note number MIDI note number 70 68 66 64 62 60 58 A part of MIDI melody Melody notes An extracted pitch vector 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 ...