Music Database Research Articles

Identyfikacja koderów MP3 oraz urządzeń rejestrujących na podstawie badania plików fonicznych poddanych stratnej kompresji

The paper outlines the problem of identification of MP3 encoders and recording devices based on the analysis of audio recordings subjected to lossy date compression. The proposed method can be used as a support for other solutions used to detect double compression and discontinuities. The approach is based on the statistical analysis of the variables obtained directly from the MP3 data stream and constitute an inherent element of compression performance. Designated vectors consisting of 46 features were used as training sequences of linear discriminant analysis (LDA), one of the most popular supervised machine learning algorithms. The effectiveness of this algorithms for the identification of MP3 encoders and recording equipment was tested on a music database consisting of nearly one million MP3 files, specially prepared for this purpose. The results of the research were discussed in the context of the influence of compression parameters on the ability to detect falsification in digital audio recordings.

BaNa: A Noise Resilient Fundamental Frequency Detection Algorithm for Speech and Music

Fundamental frequency (F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sub> ) is one of the essential features in many acoustic related applications. Although numerous F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sub> detection algorithms have been developed, the detection accuracy in noisy environments still needs improvement. We present a hybrid noise resilient F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sub> detection algorithm named BaNa that combines the approaches of harmonic ratios and Cepstrum analysis. A Viterbi algorithm with a cost function is used to identify the F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sub> value among several F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sub> candidates. Speech and music databases with eight different types of additive noise are used to evaluate the performance of the BaNa algorithm and several classic and state-of-the-art F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sub> detection algorithms. Results show that for almost all types of noise and signal-to-noise ratio (SNR) values investigated, BaNa achieves the lowest Gross Pitch Error (GPE) rate among all the algorithms. Moreover, for the 0 dB SNR scenarios, the BaNa algorithm is shown to achieve 20% to 35% GPE rate for speech and 12% to 39% GPE rate for music. We also describe implementation issues that must be addressed to run the BaNa algorithm as a real-time application on a smartphone platform.

Encoding medieval music notation for research

In 1984, John Stinson and Brian Parish developed Scribe, a computer program to encode every meaningful mark on each page of a medieval music manuscript and produce an on-screen representation of these data in both medieval and modern notation. Scribe data have proved essential for creating statistical and comparative analyses, compositional analyses and producing online thematic indices for the Medieval Music Database over a large body of music. Even though the Scribe still functions in cross-platform DOS-emulated computer environments, the growth of Digital Humanities, linked open data and enormous potential for online research collaboration offers a series of opportunities for encoded medieval music notation data. This report details the authors’ efforts since 2013 in converting Scribe’s data into open access data based upon the standard being developed by the Music Encoding Initiative (MEI). When coupled with recent developments in the Standard Music Font Layout (SMuFL) project, our new Scribe-based module, known as NeoScribe, offers significant enhancements to the MEI standard that stand to benefit current and future developments in digital musicology.

Progressive Filtering Approach for Query by Humming System Through Empirical Mode Decomposition and Multiresolution Histograms

AbstractThis research work proposes an implementation of adept content-based music retrieval technique that attempts to address the demands of the rising availability of digital music. The primary objective of this research work is to balance the perilous impact of non-relevant songs through progressive filtering (PF) for query by humming (QBH) music information retrieval system. The PF is a technique of searching in manifolds for problem solving through reduced search space. A new strategy for empirical mode decomposition (EMD) analysis is adopted, and outcomes are propelled as a significant source of information for multiresolution histogram (MRH) representations. Finally, the concept of PF is realized through MRH to retrieve the desired song by humming. The proposed multiresolution QBH system is evaluated on a music database of 1200 fragments, and the experimental results effectively locate the humming patterns and recite the eminence of EMD analysis for retrieving the desired song through humming.

Query by humming based on multiple spectral hashing and scaled open-end dynamic time warping

Query by humming (QBH) is to retrieve songs in the music database by using user׳s humming. In QBH, the huge size of a song database requires an efficient search method. Recently, local sensitive hashing (LSH) has been applied in QBH and showed its superior performance. In this paper, we propose a method for QBH which uses multiple spectral hashing (MSH) and scaled open-end dynamic time warping (SOEDTW). We construct multiple binary embedding spaces by utilizing eigenvectors obtained from spectral hashing, so we call this approach as multiple spectral hashing (MSH). We also apply an improved OEDTW method for similarity matching. The experimental results demonstrate that the proposed method can improve retrieval performance greatly.

“New Possibility of Research on East Asian Music through Establishment of Music Database of Primary Sources”

“New Possibility of Research on East Asian Music through Establishment of Music Database of Primary Sources”

Music information retrieval in compressed audio files: a survey

In this paper, we present an organized survey of the existing literature on music information retrieval systems in which descriptor features are extracted directly from the compressed audio files, without prior decompression to pulse-code modulation format. Avoiding the decompression step and utilizing the readily available compressed-domain information can significantly lighten the computational cost of a music information retrieval system, allowing application to large-scale music databases. We identify a number of systems relying on compressed-domain information and form a systematic classification of the features they extract, the retrieval tasks they tackle and the degree in which they achieve an actual increase in the overall speed—as well as any resulting loss in accuracy. Finally, we discuss recent developments in the field, and the potential research directions they open toward ultra-fast, scalable systems.

Authenticity examination of compressed audio recordings using detection of multiple compression and encoders’ identification

Since the appearance of digital audio recordings, audio authentication has been becoming increasingly difficult. The currently available technologies and free editing software allow a forger to cut or paste any single word without audible artifacts. Nowadays, the only method referring to digital audio files commonly approved by forensic experts is the ENF criterion. It consists in fluctuation analysis of the mains frequency induced in electronic circuits of recording devices. Therefore, its effectiveness is strictly dependent on the presence of mains signal in the recording, which is a rare occurrence. Recently, much attention has been paid to authenticity analysis of compressed multimedia files and several solutions were proposed for detection of double compression in both digital video and digital audio. This paper addresses the problem of tampering detection in compressed audio files and discusses new methods that can be used for authenticity analysis of digital recordings. Presented approaches consist in evaluation of statistical features extracted from the MDCT coefficients as well as other parameters that may be obtained from compressed audio files. Calculated feature vectors are used for training selected machine learning algorithms. The detection of multiple compression covers up tampering activities as well as identification of traces of montage in digital audio recordings. To enhance the methods’ robustness an encoder identification algorithm was developed and applied based on analysis of inherent parameters of compression. The effectiveness of tampering detection algorithms is tested on a predefined large music database consisting of nearly one million of compressed audio files. The influence of compression algorithms’ parameters on the classification performance is discussed, based on the results of the current study.

Directed Acyclic Graphs for Content Based Sound, Musical Genre, and Speech Emotion Classification

This work introduces the methodology of Decision Directed Acyclic Graphs (DDAG)11 This work uses the following abbreviations as regards to directed graphs: Directed Acyclic Graph (DAG), Decision Directed Acyclic Graph (DDAG) and Directed Acyclic Graph Hidden Markov Model (DAGHMM). to the scientific domain of content based audio signal processing. We apply the particular methodology to three multiclass classification problems involving the categories of generalized sound events, musical genres, and speech expressing emotional states. A decision graph is constructed which breaks the overall problem into a series of two-class ones. The order of the graph nodes is revealed using a clustering criterion based on the Kullback-Leibler divergence. Every graph node is composed by two hidden Markov models, each one representing the class which participates in the specific problem. We extract three heterogeneous feature sets (Mel-Filterbank, MPEG-7 Audio Spectrum Projection and Perceptual Wavelet Packets) out of each recording and fuse them for training the HMMs. Extensive comparative experiments are conducted using the following three datasets: (a) a combination of professional sound effects collections, (b) GTZAN musical genre database, and (c) BERLIN emotional speech corpus. The results demonstrate the superiority of the DDAG classification approach over the standard HMM approach regardless the application task.

Emotion Recognition in Music Signal using AANN and SVM

main objective of this work is to develop a music emotion recognition technique using Mel frequency cepstral coefficient (MFCC), Auto associative neural network (AANN) and support vector machine (SVM). The emotions taken are anger, happy, sad, fear, and neutral. Music database is collected at 44.1 KHz with 16 bits per sample from various movies and websites related to music. For each emotion 15 music signals are recorded and each one is by 15sec duration. The proposed technique of music emotion recognition (MER) is done in two phases such, i) Feature extraction, and ii) Classification. Initially, music signal is given to feature extraction phase to extract MFCC features. Second the extracted features are given to Auto associative neural networks (AANN) and support vector machine (SVM) classifiers to categorize the emotions and finally their performance are compared. The experimental results show that MFCC with AANN classifier achieves a recognition rate of about 94.4% and with SVM classifier of about 85.0% thus outperforms SVM classifier.

The Cultural Impact of Institutional Remix: The Formalisation of Textual Reappropriation within the ABC

In 2009, Mark Deuze proposed an updated approach to media studies to incorporate ‘media life’, a concept he suggests addresses the invisibleness of ubiquitous media. Media life provides a useful lens for researchers to understand the human condition in media and not with media. At a similar time, public service media (PSM) strategies have aligned audience participation with the so‐called Reithian trinity which suggest the PSB should inform, educate and entertain while performing its core values of public service broadcasting (Enli 2008). Remix within the PSM institution relies on audience participation, employing ‘the people formerly known as the audience’ (Rosen 2006) as cultural artifact producers, and draws on their experience from within the media. Remix as a practice then enables us to examine the shift of the core PSM values by understanding how audience participation, informed by a human condition mobilised from our existence of being in media and not merely with media. However, remix within PSM challenges the once elitist construction of meaning models with an egalitarian approach towards socially reappropriated texts, questioning its affect on the cultural landscape. This paper draws on three years of ethnographic data from within the Australian Broadcasting Corporation (ABC), exploring the remix culture of ABC Pool. ABC Pool operates under a Creative Commons licensing regime to enable remix practice under the auspices of the ABC. ABC Pool users provide a useful group of remix practitioners to examine as they had access to a vast ABC archival collection and were invited to remix those cultural artefacts, often adding cultural and fiscal value. This paper maintains a focus on the audience participation within PSM through remix culture by applying media dependency theory to remix as cultural practice and calls to expand and update the societal representation within the ABC.

Low-order auditory Zernike moment: a novel approach for robust music identification in the compressed domain

Audio identification via fingerprint has been an active research field for years. However, most previously reported methods work on the raw audio format in spite of the fact that nowadays compressed format audio, especially MP3 music, has grown into the dominant way to store music on personal computers and/or transmit it over the Internet. It will be interesting if a compressed unknown audio fragment could be directly recognized from the database without decompressing it into the wave format at first. So far, very few algorithms run directly on the compressed domain for music information retrieval, and most of them take advantage of the modified discrete cosine transform coefficients or derived cepstrum and energy type of features. As a first attempt, we propose in this paper utilizing compressed domain auditory Zernike moment adapted from image processing techniques as the key feature to devise a novel robust audio identification algorithm. Such fingerprint exhibits strong robustness, due to its statistically stable nature, against various audio signal distortions such as recompression, noise contamination, echo adding, equalization, band-pass filtering, pitch shifting, and slight time scale modification. Experimental results show that in a music database which is composed of 21,185 MP3 songs, a 10-s long music segment is able to identify its original near-duplicate recording, with average top-5 hit rate up to 90% or above even under severe audio signal distortions.

Automatic Music Genre Classification of Audio Signals with Machine Learning Approaches

AbstractMusical genre classification is put into context by explaining about the structures in music and how it is analyzed and perceived by humans. The increase of the music databases on the personal collection and the Internet has brought a great demand for music information retrieval, and especially automatic musical genre classification. In this research we focused on combining information from the audio signal than different sources. This paper presents a comprehensive machine learning approach to the problem of automatic musical genre classification using the audio signal. The proposed approach uses two feature vectors, Support vector machine classifier with polynomial kernel function and machine learning algorithms. More specifically, two feature sets for representing frequency domain, temporal domain, cepstral domain and modulation frequency domain audio features are proposed. Using our proposed features SVM act as strong base learner in AdaBoost, so its performance of the SVM classifier cannot improve using boosting method. The final genre classification is obtained from the set of individual results according to a weighting combination late fusion method and it outperformed the trained fusion method. Music genre classification accuracy of 78% and 81% is reported on the GTZAN dataset over the ten musical genres and the ISMIR2004 genre dataset over the six musical genres, respectively. We observed higher classification accuracies with the ensembles, than with the individual classifiers and improvements of the performances on the GTZAN and ISMIR2004 genre datasets are three percent on average. This ensemble approach show that it is possible to improve the classification accuracy by using different types of domain based audio features.

Time and spectral analysis methods with machine learning for the authentication of digital audio recordings

This paper addresses the problem of tampering detection and discusses new methods that can be used for authenticity analysis of digital audio recordings. Nowadays, the only method referred to digital audio files commonly approved by forensic experts is the ENF criterion. It consists in fluctuation analysis of the mains frequency induced in electronic circuits of recording devices. Therefore, its effectiveness is strictly dependent on the presence of mains signal in the recording, which is a rare occurrence. This article presents the existing methods of time and spectral analysis along with their modifications as proposed by the author involving spectral analysis of residual signal enhanced by machine learning algorithms. The effectiveness of tampering detection methods described in this paper is tested on a predefined music database. The results are compared graphically using ROC-like curves. Furthermore, time–frequency plots are presented and enhanced by reassignment method in purpose of visual inspection of modified recordings. Using this solution, enables analysis of minimal changes of background sounds, which may indicate tampering.

LibraryMusicSource.com (review)

Reviewed by: LibraryMusicSource.com Scott Stone LibraryMusicSource.com. [Verona, NJ]: CD Sheet Music, LLC, 2000–. www.librarymusicsource.com (Accessed July–August 2012). [Requires a Web browser, Adobe Reader, and an Internet connection. Pricing: institutional subscription starts at $500 per year for unlimited users and downloads; individual subscription for $19.95 for 40 downloads with no time limit.] Many music libraries went through a brief period of collection development during which librarians purchased compact discs containing hundreds of pages of scanned sheet music. Instead of purchasing a single printed score, a library could pay about the same amount of money for dozens of titles on one easy-to-use disc—what a great deal! Perhaps the largest of these compact disc publishers, CD Sheet Music, took the next digital step forward by offering institutions and individuals a subscription option to their entire collection. This subscription product is Library Music Source. Library Music Source (LMS) is a subscription sheet music database that offers the entire collection of CD Sheet Music and the Orchestra Musician’s CD-ROM Library. This collection amounts to more than 300,000 pages of music comprising approximately 35,000 works. The collection provides public domain sheet music for almost all instruments commonly studied in American music schools, but primarily focuses on solo piano, piano duets, solo violin, violin and piano, voice and piano, opera vocal scores, choral vocal scores, full orchestral scores, and individual instrumental parts to major orchestral works. The search box, found on every page of the site, only allows for simple keyword searches. This could sometimes be frustrating as I would occasionally retrieve far more results than I was expecting; fortunately, LMS makes up for their simple search capabilities by providing rather good browsing options. One may begin browsing by either composer’s surname or by instrument family, both of which quickly branch out into multiple options. For example, I was able to very quickly locate the [End Page 602] Concerto in D Minor, op.47 for violin by Jean Sibelius with a quick browse; unfortunately, the resulting music was not what I expected as it provided me with the piano reduction score even though the Browsing category was clearly labeled “Solo Violin with Orchestra.” While the score was not exactly what I wanted, I downloaded the PDF file with a single click and was rewarded with a very clean scan of some score. I describe it as “some,” because LMS does not provide any information as to who published the original score, which I would certainly like to know so that I could easily compare editions if I were preparing for a performance. This issue became even more problematic to me when I noticed scores that contain fingerings without any information as to who originally provided the fingerings. The CD Sheet Music Web site states that scores come from publishers such as Durand, Schirmer, Peters, Breitkopf & Härtel, and others, but they “do not identify the edition of each work” because the “collections include so many individual works from a variety of different published editions.”1 Next, I decided to look at the vocal selections, because I work with many voice students. LMS certainly has many of the standard literature by Brahms, Schubert, and Fauré that students frequently perform; however, I was unable to find both high and low voicings for all the songs I searched for. Additionally, songs were not all arranged in a standard manner for each composer. For example, all of Schubert’s songs appear to be available individually by each individual song title (and therefore downloadable as a single file), but most of Fauré’s songs are grouped by opus (and therefore one must download a large group of songs at once). Somewhat frustrated by the vocal music, I next looked at the orchestral parts. As someone who spent many hours practicing (and sometimes almost as many searching for) excerpts, I was happy to see so many standard excerpts in one place. Most of the excerpt books I’ve ever used generally provide only one or two small parts of the whole piece, but these excerpts provide an instrument’s entire part for a work so one could truly study...

Index Compression for Audio Fingerprinting Systems Based on Compressed Suffix Array

 Abstract—As one of most popular technologies, audio fingerprinting has recently attracted much attention in music retrieval systems. In music retrieval methods based on audio fingerprints, a large database is required in order to compare the fingerprints extracted from the query. In other words, the efficient search method has to be developed. In this paper, we propose a method for index compression using a compressed suffix array. Taking advantage of the fact that the repetitive characters occur frequently in higher bits of the sorted audio fingerprint data, the proposed method compresses the index by encoding the 8-bit data sequences by Run Length Encoding. Vertical Code is also used to compress the array, wherein the positions of the sorted data are stored. Four sets of music databases are used in experiments to evaluate the effectiveness of the proposed method. The experimental results show that the proposed method, compared with the conventional method, only needs 30% of the space of an audio fingerprints database for a music database consisting of 8000 songs, and around 80% of the index space for a database of 1000 songs. Moreover, the entire space cost is reduced to around 60%, compared with the method based on the suffix array. using a suffix array (6) has also been proposed. In the method based on the suffix array, the space cost increases in proportion to the growing music database. In this paper, we proposed a method to reduce the space cost by compressing the index of the database. The paper is organized as follows: Section II outlines music retrieval based on audio fingerprints. We review a fast Hamming space search method (7) based on a suffix array in Section III, and propose a space-saving method based on a compressed suffix array in Section IV. We evaluate the proposed method in Section V. Finally, the conclusions and future work are given in Section VI.

Directing Physiology and Mood through Music: Validation of an Affective Music Player

Music is important in everyday life, as it provides entertainment and influences our moods. As music is widely available, it is becoming increasingly difficult to select songs to suit our mood. An affective music player can remove this obstacle by taking a desired mood as input and then selecting songs that direct toward that desired mood. In the present study, we validate the concept of an affective music player directing the energy dimension of mood. User models were trained for 10 participants based on skin conductance changes to songs from their own music database. Based on the resulting user models, the songs that most increased or decreased the skin conductance level of the participants were selected to induce either a relatively energized or a calm mood. Experiments were conducted in a real-world office setting. The results showed that a reliable prediction can be made of the impact of a song on skin conductance, that skin conductance and mood can be directed toward an energized or calm state and that skin conductance remains in these states for at least 30 minutes. All in all, this study shows that the concept and models of the affective music player worked in an ecologically valid setting, suggesting the feasibility of using physiological responses in real-life affective computing applications.

강인한 오디오 핑거프린팅 시스템을 위한 에너지와 통계적 필터링

디지털 음악과 스마트 폰이 대중화되면서 잡음에 강인한 실시간 음악 핑거프린트 시스템이 다양하게 개발되고 있다. 특히 핑거프린트 알고리즘 중 Multiple Hashing(MLH)은 잡음에 강인하고 정교한 구조로 되어 있다. 본 논문에서는 음악 데이터베이스로부터 질의 및 응답의 정확도를 개선하기 위해 에너지 집중필터를 사용하고 연속성과 중복성을 제거하는 통계적 필터를 제안한다. 에너지 집중 필터는 하위 비트에 에너지가 집중되는 Discrite Cosine Transform(DCT)의 특징을 이용하고, 통계적 필터는 검색된 핑거프린트 정보들 사이의 상관관계 특성을 이용한다. 실험 결과로 잡음 환경에서 에너지와 통계적 필터링으로 구성된 제안 알고리즘은 우수성을 보인다. 이는 제안된 필터 엔진으로 Philips Robust Hash(PRH)보다 잡음에 강인하고 Multiple Hashing(MLH)보다 간결한 핑거프린트 시스템을 구성할 수 있다. The popularity of digital music and smart phones led to develope noise-robust real-time audio fingerprinting system in various ways. In particular, The Multiple Hashing(MLH) of fingerprint algorithms is robust to noise and has an elaborate structure. In this paper, we propose a filter engine based on MLH to achieve better performance. In this approach, we compose a energy-intensive filter to improve the accuracy of Q/R from music database and a statistic filter to remove continuity and redundancy. The energy-intensive filter uses the Discrite Cosine Transform(DCT)'s feature gathering energy to low-order bits and the statistic filters use the correlation between searched fingerprint's information. Experimental results show that the superiority of proposed algorithm consists of the energy and statistical filtering in noise environment. It is found that the proposed filter engine achieves more robust to noise than Philips Robust Hash(PRH), and a more compact way than MLH.

Music genre classification using LBP textural features

In this paper we present an approach to music genre classification which converts an audio signal into spectrograms and extracts texture features from these time-frequency images which are then used for modeling music genres in a classification system. The texture features are based on Local Binary Pattern, a structural texture operator that has been successful in recent image classification research. Experiments are performed with two well-known datasets: the Latin Music Database (LMD), and the ISMIR 2004 dataset. The proposed approach takes into account some different zoning mechanisms to perform local feature extraction. Results obtained with and without local feature extraction are compared. We compare the performance of texture features with that of commonly used audio content based features (i.e. from the MARSYAS framework), and show that texture features always outperforms the audio content based features. We also compare our results with results from the literature. On the LMD, the performance of our approach reaches about 82.33%, above the best result obtained in the MIREX 2010 competition on that dataset. On the ISMIR 2004 database, the best result obtained is about 80.65%, i.e. below the best result on that dataset found in the literature.

Data collection for individuality analysis on subjective music similarity evaluation

Recently there are many studies of subjective music similarity for music information retrieval. To quantify the subjective music similarity, there are many factors to take into account. In this study the individuality of the subjective similarity was focused on. To analyze the individuality of the subjective similarity, subjective evaluation data for 200 pairs of RWC popular music database was collected. 28 subjects listened to the pairs of tracks, and evaluated similarity of each pair by similar or dissimilar. They also selected the components of music (melody, tempo/rhythm, vocal, instruments) that was similar. Each subject evaluated the same 200 pairs, thus the individuality of the evaluation can be easily analyzed. The results of pilot analysis were reported.