Towards Cross-Version Singing Voice Detection

National Conference on Signal and Image Processing Applications

Music information retrieval is currently an active research area that addresses the extraction of musically important information from audio signals, and the applications of such information. The extracted information can be used for search and retrieval of music in recommendation systems, or to aid musicological studies or even in music learning. Sophisticated signal processing techniques are applied to convert low-level acoustic signal properties to musical attributes which are further embedded in a rulebased or statistical classification framework to link with high-level descriptions such as melody, genre, mood and artist type. Vocal music comprises a large and interesting category of music where the lead instrument is the singing voice. The singing voice is more versatile than many musical instruments and therefore poses interesting challenges to information retrieval systems. In this paper, we provide a brief overview of research in vocal music processing followed by a description of related work at IIT Bombay leading to the development of an interface for melody detection of singing voice in polyphony.

Entropy, 2022

Singing voice detection or vocal detection is a classification task that determines whether there is a singing voice in a given audio segment. This process is a crucial preprocessing step that can be used to improve the performance of other tasks such as automatic lyrics alignment, singing melody transcription, singing voice separation, vocal melody extraction, and many more. This paper presents a survey on the techniques of singing voice detection with a deep focus on state-of-the-art algorithms such as convolutional LSTM and GRU-RNN. It illustrates a comparison between existing methods for singing voice detection, mainly based on the Jamendo and RWC datasets. Long-term recurrent convolutional networks have reached impressive results on public datasets. The main goal of the present paper is to investigate both classical and state-of-the-art approaches to singing voice detection.

2009

Detecting distinct features in modern pop music is an important problem that can have significant applications in areas such as multimedia entertainment. They can be used, for example, to give a visually coherent representation of the sound. We propose to integrate a singing voice detector with a multimedia, multi-touch game where the user has to perform simple tasks at certain key points in the music. While the ultimate goal is to automatically create visual content in response to features extracted from the music, here we give special focus to the detection of voice segments in music songs. The solution presented extracts the Mel-Frequency Cepstral Coefficients of the sound and uses a Hidden Markov Model to infer if the sound has voice. The classification rate obtained is high when compared to other singing voice detectors that use Mel-Frequency Cepstral Coefficients.

2009

In this article, we present an improvement of a previous singing voice detector. This new detector is in two steps.

2012

This dissertation is concerned with the problem of describing the singing voice within the audio signal of a song. This work is motivated by the fact that the lead vocal is the element that attracts the attention of most listeners. For this reason it is common for music listeners to organize and browse music collections using information related to the singing voice such as the singer name. Our research concentrates on the three major problems of music information retrieval: the localization of the source to be described (i.e. the recognition of the elements corresponding to the singing voice in the signal of a mixture of instruments), the search of pertinent features to describe the singing voice, and finally the development of pattern recognition methods based on these features to identify the singer. For this purpose we propose a set of novel features computed on the temporal variations of the fundamental frequency of the sung melody. These features, which aim to describe the vib...

Automatic singing detection and singing phoneme recognition are two MIR research topics that have gained a lot of attention the last years. The,rst approaches borrowed successful techniques widely used in Automatic Speech Recognition (ASR) as speech and singing share similar acoustical features since they are produced by the same ap- paratus. Moving from monophonic to polyphonic audio signals the problem becomes more complex as the background instrumental ac- companiment is regarded as a noise source that has to be attenuated. This thesis presents research into the problem of singing phoneme de- tection in polyphonic audio, in which the lyrics are in English. Speci- cally, we are interested in building statistical classication models that are able to automatically distinguish sung consonants and vowels from pure instrumental music in polyphonic music recordings.

IEEE Transactions on Audio, Speech and Language Processing, 2000

In this paper, we investigate the problem of automatic singer identification, detection and tracking in popular music recordings with one or multiple singers. This problem reflects an important issue in multimedia applications that require the transcription and indexing of music data to meet the increasing demand for content-based information retrieval. The major challenges for this study arise from the fact that a singer's voice tends to be arbitrarily altered from time to time and is inextricably intertwined with the signal of the background accompaniment. To determine who is singing, or whether or when a particular singer is present in a music recording, methods are presented for separating vocal from non-vocal regions, for isolating singers' vocal characteristics from background music, and for distinguishing singers from one another. Experimental evaluations conducted on a pop music database consisting of solo and duet tracks confirm the validity of the proposed methods.

International Journal for Research in Applied Science and Engineering Technology

I. INTRODUCTION This study introduces a technique for automatically identifying singers that groups songs based on voice resemblance after analysing musical signals. The technique employs a statistical model to distinguish singing voices from other sounds and identify singers. With an accuracy rate of 80%, it extracts audio components and matches them with singer voice models. To evaluate vocalist identification strategies in polyphonic music, the method additionally makes use of pattern classification and vocal separation algorithms. The project also investigates the use of timbre data to detect vocal qualities of vocalists and middle-level variables to capture perceptual properties of music. On the Artist20 benchmark dataset, the proposed convolutional recurrent neural network (CRNN) achieves an average F1 score of 0.81. The study looks into how popular song performers might be recognised using vibrato traits. The method computes cepstral coefficients based on vibration features and leverages high-level musical experience in song structure. The work generates polyphonic musical audio signals, models the characteristics of a singing voice, and employs MFCC and LPC coefficients to identify artists in Indian video songs. The work provides methods for modelling the characteristics of a singing voice and producing polyphonic musical audio signals. As domain adaptation strategies, the method employs gradient reversal, contrastive adaptation network (CAN), and maximum mean discrepancy (MMD). The study looks at the use of 32 Mel-Frequency Cepstral Coefficients in two subsets: low order MFCCs that represent vocal tract resonances and high order MFCCs linked to glottal wave shape. Additionally, the study looks at how polyphonic audio signals can imitate the characteristics of singing voices utilised in Sri Lankan music. A deep neural network model called the KNN-Net for SID learns the representation of local timbre features from the vocalist's voice and the background music. The authors present novel methods for learning classifiers directly from polyphonic musical signals to improve MIR systems. These techniques are meant to increase the metaverse's vocalist identification's precision and dependability. As music databases grow, solutions are needed to efficiently classify and retrieve digital music libraries. Automatic singer identification (SID), which recognises singing performers in audio samples, is an essential part of MIR systems. Traditional methods like the Hidden Markov Model (HMM), Support Vector Machine (SVM), and Gaussian Mixture Model (GMM) are not flawless. Researchers have suggested utilising Open-unmix, an open-source tool, to separate singing from accompaniment. For speech and music recognition, respectively, Mel spectrum cepstral coefficients (MFCC) and linear prediction cepstral coefficients (LPCC) are employed. To improve SID accuracy, Nasrullah and Zhao propose a novel model based on a convolutional recurrent neural network (CRNN) framework. The development of online music archives has led to the emergence of content-based music retrieval algorithms, with automatic artist identification playing a crucial role. Identifying singing segments, isolating vocal segment characteristics, and building a singer classifier using feature parameters are the three primary processes in the critical MIR problem of singer identification (SingerID). Recent studies have concentrated on perceptually motivated characteristics of singing voices for music content processing and analysis, such as vibrato and harmonic acoustic characteristics. This study uses higher-order cepstral coefficients related to pitch and fine spectral structure to characterise singing voices with a focus on individuality rather than formantic organisation. The definition of singing patterns, author classification of music, and singer automatic recognition could be the main areas of future study. II. CONCLUSION The article addresses automatic singer recognition (SAR) methods that improve accuracy and get closer to Kullback-Leibler divergence in polyphonic music and split vocals. The approach focuses on middle-level musical features and employs convolutional and recurrent networks. It offers HSI, a groundbreaking technique with extraordinary efficacy, scalability, efficiency, and durability for identifying singers in large music libraries.

Applied Sciences

There are insufficient datasets of singing files that are adequately annotated. One of the available datasets that includes a variety of vocal techniques (n = 17) and several singers (m = 20) with several WAV files (p = 3560) is the VocalSet dataset. However, although several categories, including techniques, singers, tempo, and loudness, are in the dataset, they are not annotated. Therefore, this study aims to annotate VocalSet to make it a more powerful dataset for researchers. The annotations generated for the VocalSet audio files include fundamental frequency contour, note onset, note offset, the transition between notes, note F0, note duration, Midi pitch, and lyrics. This paper describes the generated dataset and explains our approaches to creating and testing the annotations. Moreover, four different methods to define the onset/offset are compared.

2005 IEEE International Conference on Multimedia and Expo

In this work we strive to find an optimal set of acoustic features for the discrimination of speech, monophonic singing, and polyphonic music to robustly segment acoustic media streams for annotation and interaction purposes. Furthermore we introduce ensemble-based classification approaches within this task. From a basis of 276 attributes we select the most efficient set by SVM-SFFS. Additionally relevance of single features by calculation of information gain ratio is presented. As a basis of comparison we reduce dimensionality by PCA. We show extensive analysis of different classifiers within the named task. Among these are Kernel Machines, Decision Trees, and Bayesian Classifiers. Moreover we improve single classifier performance by Bagging and Boosting, and finally combine strengths of classifiers by StackingC. The database is formed by 2,114 samples of speech, and singing of 58 persons. 1,000 Music clips have been taken from the MTV-Europe-Top-20 1980-2000. The outstanding discrimination results of a working realtime capable implementation stress the practicability of the proposed novel ideas.