Novel melody line identification algorithm for polyphonic MIDI music

The problem of automatic melody line identification in a MIDI file plays an important role towards taking QBH systems to the next level. We present here, a novel algorithm to identify the melody line in a polyphonic MIDI file. A note pruning and track/channel ranking method is used to identify the melody line. We use results from musicology to derive certain simple heuristics for the note pruning stage. This helps in the robustness of the algorithm, by way of discarding "spurious" notes. A ranking based on the melodic information in each track/channel enables us to choose the melody line accurately. Our algorithm makes no assumption about MIDI performer specific parameters, is simple and achieves an accuracy of 97% in identifying the melody line correctly. This algorithm is currently being used by us in a QBH system built in our lab.

A speech-music discriminator using HILN model based features

We propose a simple speech music discriminator that uses features based on HILN(Harmonics, Individual Lines and Noise) model. We have been able to test the strength of the feature set on a standard database of 66 files and get an accuracy of around 97%. We also have tested on sung queries and polyphonic music and have got very good results. The current algorithm is being used to discriminate between sung queries and played (using an instrument like flute) queries for a Query by Humming(QBH) system currently under development in the lab.

Finite data performance of MUSIC and minimum norm methods

In the direction of arrival (DOA) estimation problem, we encounter both finite data and insufficient knowledge of array characterization. It is therefore important to study how subspace-based methods perform in such conditions. We analyze the finite data performance of the multiple signal classification (MUSIC) and minimum norm (min. norm) methods in the presence of sensor gain and phase errors, and derive expressions for the mean square error (MSE) in the DOA estimates. These expressions are first derived assuming an arbitrary array and then simplified for the special case of an uniform linear array with isotropic sensors. When they are further simplified for the case of finite data only and sensor errors only, they reduce to the recent results given in [9-12]. Computer simulations are used to verify the closeness between the predicted and simulated values of the MSE.

Carnatic music analysis: Shadja, swara identification and rAga verification in AlApana using stochastic models

We analyze the AlApana of a Carnatic music piece without the prior knowledge of the singer or the rAga. AlApana is ameans to communicate to the audience, the flavor or the bhAva of the rAga through the permitted notes and its phrases. The input to our analysis is a recording of the vocal AlApana along with the accompanying instrument. The AdhAra shadja(base note) of the singer for that AlApana is estimated through a stochastic model of note frequencies. Based on the shadja, we identify the notes (swaras) used in the AlApana using a semi-continuous GMM. Using the probabilities of each note interval, we recognize swaras of the AlApana. For sampurNa rAgas, we can identify the possible rAga, based on the swaras. We have been able to achieve correct shadja identification, which is crucial to all further steps, in 88.8% of 55 AlApanas. Among them (48 AlApanas of 7 rAgas), we get 91.5% correct swara identification and 62.13% correct R (rAga) accuracy.

Compressive sensing for music signals: comparison of transforms with coherent dictionaries

Compressive Sensing (CS) is a new sensing paradigm which permits sampling of a signal at its intrinsic information rate which could be much lower than Nyquist rate, while guaranteeing good quality reconstruction for signals sparse in a linear transform domain. We explore the application of CS formulation to music signals. Since music signals comprise of both tonal and transient nature, we examine several transforms such as discrete cosine transform (DCT), discrete wavelet transform (DWT), Fourier basis and also non-orthogonal warped transforms to explore the effectiveness of CS theory and the reconstruction algorithms. We show that for a given sparsity level, DCT, overcomplete, and warped Fourier dictionaries result in better reconstruction, and warped Fourier dictionary gives perceptually better reconstruction. “MUSHRA” test results show that a moderate quality reconstruction is possible with about half the Nyquist sampling.

Boosting-based transfer learning for multi-view head-pose classification from surveillance videos

This work proposes a boosting-based transfer learning approach for head-pose classification from multiple, low-resolution views. Head-pose classification performance is adversely affected when the source (training) and target (test) data arise from different distributions (due to change in face appearance, lighting, etc). Under such conditions, we employ Xferboost, a Logitboost-based transfer learning framework that integrates knowledge from a few labeled target samples with the source model to effectively minimize misclassifications on the target data. Experiments confirm that the Xferboost framework can improve classification performance by up to 6%, when knowledge is transferred between the CLEAR and FBK four-view headpose datasets.

Music and symbolic dynamics: the science behind an art

Music signals comprise of atomic notes drawn from a musical scale. The creation of musical sequences often involves splicing the notes in a constrained way resulting in aesthetically appealing patterns. We develop an approach for music signal representation based on symbolic dynamics by translating the lexicographic rules over a musical scale to constraints on a Markov chain. This source representation is useful for machine based music synthesis, in a way, similar to a musician producing original music. In order to mathematically quantify user listening experience, we study the correlation between the max-entropic rate of a musical scale and the subjective aesthetic component. We present our analysis with examples from the south Indian classical music system.

STUDENT'S-t MIXTURE MODEL BASED MULTI-INSTRUMENT RECOGNITION IN POLYPHONIC MUSIC

We address the problem of multi-instrument recognition in polyphonic music signals. Individual instruments are modeled within a stochastic framework using Student's-t Mixture Models (tMMs). We impose a mixture of these instrument models on the polyphonic signal model. No a priori knowledge is assumed about the number of instruments in the polyphony. The mixture weights are estimated in a latent variable framework from the polyphonic data using an Expectation Maximization (EM) algorithm, derived for the proposed approach. The weights are shown to indicate instrument activity. The output of the algorithm is an Instrument Activity Graph (IAG), using which, it is possible to find out the instruments that are active at a given time. An average F-ratio of 0 : 7 5 is obtained for polyphonies containing 2-5 instruments, on a experimental test set of 8 instruments: clarinet, flute, guitar, harp, mandolin, piano, trombone and violin.

Automatic Tonic Identification in Indian Art Music: Approaches and Evaluation

The tonic is a fundamental concept in Indian art music. It is the base pitch, which an artist chooses in order to construct the melodies during a rg(a) rendition, and all accompanying instruments are tuned using the tonic pitch. Consequently, tonic identification is a fundamental task for most computational analyses of Indian art music, such as intonation analysis, melodic motif analysis and rg recognition. In this paper we review existing approaches for tonic identification in Indian art music and evaluate them on six diverse datasets for a thorough comparison and analysis. We study the performance of each method in different contexts such as the presence/absence of additional metadata, the quality of audio data, the duration of audio data, music tradition (Hindustani/Carnatic) and the gender of the singer (male/female). We show that the approaches that combine multi-pitch analysis with machine learning provide the best performance in most cases (90% identification accuracy on average), and are robust across the aforementioned contexts compared to the approaches based on expert knowledge. In addition, we also show that the performance of the latter can be improved when additional metadata is available to further constrain the problem. Finally, we present a detailed error analysis of each method, providing further insights into the advantages and limitations of the methods.

Activity Recognition in Egocentric Life-Logging Videos

With the increasing availability of wearable cameras, research on first-person view videos (egocentric videos) has received much attention recently. While some effort has been devoted to collecting various egocentric video datasets, there has not been a focused effort in assembling one that could capture the diversity and complexity of activities related to life-logging, which is expected to be an important application for egocentric videos. In this work, we first conduct a comprehensive survey of existing egocentric video datasets. We observe that existing datasets do not emphasize activities relevant to the life-logging scenario. We build an egocentric video dataset dubbed LENA (Life-logging EgoceNtric Activities) (http://people.sutd.edu.sg/similar to 1000892/dataset) which includes egocentric videos of 13 fine-grained activity categories, recorded under diverse situations and environments using the Google Glass. Activities in LENA can also be grouped into 5 top-level categories to meet various needs and multiple demands for activities analysis research. We evaluate state-of-the-art activity recognition using LENA in detail and also analyze the performance of popular descriptors in egocentric activity recognition.

Sparse Representation based Anomaly Detection using HOMV in H.264 Compressed Videos

In this paper, we have proposed an anomaly detection algorithm based on Histogram of Oriented Motion Vectors (HOMV) 1] in sparse representation framework. Usual behavior is learned at each location by sparsely representing the HOMVs over learnt normal feature bases obtained using an online dictionary learning algorithm. In the end, anomaly is detected based on the likelihood of the occurrence of sparse coefficients at that location. The proposed approach is found to be robust compared to existing methods as demonstrated in the experiments on UCSD Ped1 and UCSD Ped2 datasets.

MULTI-INSTRUMENT DETECTION IN POLYPHONIC MUSIC USING GAUSSIAN MIXTURE BASED FACTORIAL HMM

We formulate the problem of detecting the constituent instruments in a polyphonic music piece as a joint decoding problem. From monophonic data, parametric Gaussian Mixture Hidden Markov Models (GM-HMM) are obtained for each instrument. We propose a method to use the above models in a factorial framework, termed as Factorial GM-HMM (F-GM-HMM). The states are jointly inferred to explain the evolution of each instrument in the mixture observation sequence. The dependencies are decoupled using variational inference technique. We show that the joint time evolution of all instruments' states can be captured using F-GM-HMM. We compare performance of proposed method with that of Student's-t mixture model (tMM) and GM-HMM in an existing latent variable framework. Experiments on two to five polyphony with 8 instrument models trained on the RWC dataset, tested on RWC and TRIOS datasets show that F-GM-HMM gives an advantage over the other considered models in segments containing co-occurring instruments.