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.

Monte Carlo filtering and smoothing with application to time-varying spectral estimation

We develop methods for performing filtering and smoothing in non-linear non-Gaussian dynamical models. The methods rely on a particle cloud representation of the filtering distribution which evolves through time using importance sampling and resampling ideas. In particular, novel techniques are presented for generation of random realisations from the joint smoothing distribution and for MAP estimation of the state sequence. Realisations of the smoothing distribution are generated in a forward-backward procedure, while the MAP estimation procedure can be performed in a single forward pass of the Viterbi algorithm applied to a discretised version of the state space. An application to spectral estimation for time-varying autoregressions is described.

Asymptotic behaviour of approximate Bayesian estimators

Although approximate Bayesian computation (ABC) has become a popular technique for performing parameter estimation when the likelihood functions are analytically intractable there has not as yet been a complete investigation of the theoretical properties of the resulting estimators. In this paper we give a theoretical analysis of the asymptotic properties of ABC based parameter estimators for hidden Markov models and show that ABC based estimators satisfy asymptotically biased versions of the standard results in the statistical literature.

Statistical user simulation with a hidden agenda

Recent work in the area of probabilistic user simulation for training statistical dialogue managers has investigated a new agenda-based user model and presented preliminary experiments with a handcrafted model parameter set. Training the model on dialogue data is an important next step, but non-trivial since the user agenda states are not observable in data and the space of possible states and state transitions is intractably large. This paper presents a summary-space mapping which greatly reduces the number of state transitions and introduces a tree-based method for representing the space of possible agenda state sequences. Treating the user agenda as a hidden variable, the forward/backward algorithm can then be successfully applied to iteratively estimate the model parameters on dialogue data. © 2007 Association for Computational Linguistics.

Statistical parametric speech synthesis based on speaker and language factorization

An increasingly common scenario in building speech synthesis and recognition systems is training on inhomogeneous data. This paper proposes a new framework for estimating hidden Markov models on data containing both multiple speakers and multiple languages. The proposed framework, speaker and language factorization, attempts to factorize speaker-/language-specific characteristics in the data and then model them using separate transforms. Language-specific factors in the data are represented by transforms based on cluster mean interpolation with cluster-dependent decision trees. Acoustic variations caused by speaker characteristics are handled by transforms based on constrained maximum-likelihood linear regression. Experimental results on statistical parametric speech synthesis show that the proposed framework enables data from multiple speakers in different languages to be used to: train a synthesis system; synthesize speech in a language using speaker characteristics estimated in a different language; and adapt to a new language. © 2012 IEEE.

A backward-simulation based Rao-Blackwellized particle smoother for conditionally linear Gaussian models

In this article, we develop a new Rao-Blackwellized Monte Carlo smoothing algorithm for conditionally linear Gaussian models. The algorithm is based on the forward-filtering backward-simulation Monte Carlo smoother concept and performs the backward simulation directly in the marginal space of the non-Gaussian state component while treating the linear part analytically. Unlike the previously proposed backward-simulation based Rao-Blackwellized smoothing approaches, it does not require sampling of the Gaussian state component and is also able to overcome certain normalization problems of two-filter smoother based approaches. The performance of the algorithm is illustrated in a simulated application. © 2012 IFAC.

Bayesian non-parametrics and the probabilistic approach to modelling.

Modelling is fundamental to many fields of science and engineering. A model can be thought of as a representation of possible data one could predict from a system. The probabilistic approach to modelling uses probability theory to express all aspects of uncertainty in the model. The probabilistic approach is synonymous with Bayesian modelling, which simply uses the rules of probability theory in order to make predictions, compare alternative models, and learn model parameters and structure from data. This simple and elegant framework is most powerful when coupled with flexible probabilistic models. Flexibility is achieved through the use of Bayesian non-parametrics. This article provides an overview of probabilistic modelling and an accessible survey of some of the main tools in Bayesian non-parametrics. The survey covers the use of Bayesian non-parametrics for modelling unknown functions, density estimation, clustering, time-series modelling, and representing sparsity, hierarchies, and covariance structure. More specifically, it gives brief non-technical overviews of Gaussian processes, Dirichlet processes, infinite hidden Markov models, Indian buffet processes, Kingman's coalescent, Dirichlet diffusion trees and Wishart processes.

Efficient decoding with generative score-spaces using the expectation semiring

State-of-the-art speech recognisers are usually based on hidden Markov models (HMMs). They model a hidden symbol sequence with a Markov process, with the observations independent given that sequence. These assumptions yield efficient algorithms, but limit the power of the model. An alternative model that allows a wide range of features, including word- and phone-level features, is a log-linear model. To handle, for example, word-level variable-length features, the original feature vectors must be segmented into words. Thus, decoding must find the optimal combination of segmentation of the utterance into words and word sequence. Features must therefore be extracted for each possible segment of audio. For many types of features, this becomes slow. In this paper, long-span features are derived from the likelihoods of word HMMs. Derivatives of the log-likelihoods, which break the Markov assumption, are appended. Previously, decoding with this model took cubic time in the length of the sequence, and longer for higher-order derivatives. This paper shows how to decode in quadratic time. © 2013 IEEE.

Keyword spotting system based on biomimetic pattern recognition

The mandarin keyword spotting system was investigated, and a new approach was proposed based on the principle of homology continuity and point location analysis in high-dimensional space geometry theory which are both parts of biomimetic pattern recognition theory. This approach constructed a hyper-polyhedron with sample points in the training set and calculated the distance between each test point and the hyper-polyhedron. The classification resulted from the value of those distances. The approach was tested by a speech database which was created by ourselves. The performance was compared with the classic HMM approach and the results show that the new approach is much better than HMM approach when the training data is not sufficient.

A speaker-independent continuous speech recognition system using biomimetic pattern recognition

In speaker-independent speech recognition, the disadvantage of the most diffused technology (HMMs, or Hidden Markov models) is not only the need of many more training samples, but also long train time requirement. This paper describes the use of Biomimetic pattern recognition (BPR) in recognizing some mandarin continuous speech in a speaker-independent manner. A speech database was developed for the course of study. The vocabulary of the database consists of 15 Chinese dish's names, the length of each name is 4 Chinese words. Neural networks (NNs) based on Multi-weight neuron (MWN) model are used to train and recognize the speech sounds. The number of MWN was investigated to achieve the optimal performance of the NNs-based BPR. This system, which is based on BPR and can carry out real time recognition reaches a recognition rate of 98.14% for the first option and 99.81% for the first two options to the persons from different provinces of China speaking common Chinese speech. Experiments were also carried on to evaluate Continuous density hidden Markov models (CDHMM), Dynamic time warping (DTW) and BPR for speech recognition. The Experiment results show that BPR outperforms CDHMM and DTW especially in the cases of samples of a finite size.

Biomimetic pattern recognition for speaker-independent speech recognition

In speaker-independent speech recognition, the disadvantage of the most diffused technology ( Hidden Markov Models) is not only the need of many more training samples, but also long train time requirement. This paper describes the use of Biomimetic Pattern Recognition (BPR) in recognizing some Mandarin Speech in a speaker-independent manner. The vocabulary of the system consists of 15 Chinese dish's names. Neural networks based on Multi-Weight Neuron (MWN) model are used to train and recognize the speech sounds. Experimental results are presented to show that the system, which can carry out real time recognition of the persons from different provinces speaking common Chinese speech, outperforms HMMs especially in the cases of samples of a finite size.

超参数控制下的波阻抗约束反演

The seismic survey is the most effective geophysical method during exploration and development of oil/gas. As a main means in processing and interpreting seismic data, impedance inversion takes up a special position in seismic survey. This is because the impedance parameter is a ligament which connects seismic data with well-logging and geological information, while it is also essential in predicting reservoir properties and sand-body. In fact, the result of traditional impedance inversion is not ideal. This is because the mathematical inverse problem of impedance is poor-pose so that the inverse result has instability and multi-result, so it is necessary to introduce regularization. Most simple regularizations are presented in existent literature, there is a premise that the image(or model) is globally smooth. In fact, as an actual geological model, it not only has made of smooth region but also be separated by the obvious edge, the edge is very important attribute of geological model. It's difficult to preserve these characteristics of the model and to avoid an edge too smooth to clear. Thereby, in this paper, we propose a impedance inverse method controlled by hyperparameters with edge-preserving regularization, the inverse convergence speed and result would be improved. In order to preserve the edge, the potential function of regularization should satisfy nine conditions such as basic assumptions edge preservation and convergence assumptions etc. Eventually, a model with clear background and edge-abnormity can be acquired. The several potential functions and the corresponding weight functions are presented in this paper. The potential functionφLφHL andφGM can meet the need of inverse precision by calculating the models. For the local constant planar and quadric models, we respectively present the neighborhood system of Markov random field corresponding to the regularization term. We linearity nonlinear regularization by using half-quadratic regularization, it not only preserve the edge, and but also simplify the inversion, and can use some linear methods. We introduced two regularization parameters (or hyperparameters) λ2 and δ in the regularization term. λ2 is used to balance the influence between the data term and the transcendental term; δ is a calibrating parameter used to adjust the gradient value at the discontinuous position(or formation interface). Meanwhile, in the inverse procedure, it is important to select the initial value of hyperparameters and to change hyperparameters, these will then have influence on convergence speed and inverse effect. In this paper, we roughly give the initial value of hyperparameters by using a trend- curve of φ-(λ2, δ) and by a method of calculating the upper limit value of hyperparameters. At one time, we change hyperparameters by using a certain coefficient or Maximum Likelihood method, this can be simultaneously fulfilled with the inverse procedure. Actually, we used the Fast Simulated Annealing algorithm in the inverse procedure. This method overcame restrictions from the local extremum without depending on the initial value, and got a global optimal result. Meanwhile, we expound in detail the convergence condition of FSA, the metropolis receiving probability form Metropolis-Hasting, the thermal procession based on the Gibbs sample and other methods integrated with FSA. These content can help us to understand and improve FSA. Through calculating in the theoretic model and applying it to the field data, it is proved that the impedance inverse method in this paper has the advantage of high precision practicability and obvious effect.

地震反射系数成像方法研究

Reflectivity sequences extraction is a key part of impedance inversion in seismic exploration. Although many valid inversion methods exist, with crosswell seismic data, the frequency brand of seismic data can not be broadened to satisfy the practical need. It is an urgent problem to be solved. Pre-stack depth migration which developed in these years becomes more and more robust in the exploration. It is a powerful technology of imaging to the geological object with complex structure and its final result is reflectivity imaging. Based on the reflectivity imaging of crosswell seismic data and wave equation, this paper completed such works as follows: Completes the workflow of blind deconvolution, Cauchy criteria is used to regulate the inversion(sparse inversion). Also the precondition conjugate gradient(PCG) based on Krylov subspace is combined with to decrease the computation, improves the speed, and the transition matrix is not necessary anymore be positive and symmetric. This method is used to the high frequency recovery of crosswell seismic section and the result is satisfactory. Application of rotation transform and viterbi algorithm in the preprocess of equation prestack depth migration. In equation prestack depth migration, the grid of seismic dataset is required to be regular. Due to the influence of complex terrain and fold, the acquisition geometry sometimes becomes irregular. At the same time, to avoid the aliasing produced by the sparse sample along the on-line, interpolation should be done between tracks. In this paper, I use the rotation transform to make on-line run parallel with the coordinate, and also use the viterbi algorithm to complete the automatic picking of events, the result is satisfactory. 1. Imaging is a key part of pre-stack depth migration besides extrapolation. Imaging condition can influence the final result of reflectivity sequences imaging greatly however accurate the extrapolation operator is. The author does migration of Marmousi under different imaging conditions. And analyzes these methods according to the results. The results of computation show that imaging condition which stabilize source wave field and the least-squares estimation imaging condition in this paper are better than the conventional correlation imaging condition. The traditional pattern of "distributed computing and mass decision" is wisely adopted in the field of seismic data processing and becoming an obstacle of the promoting of the enterprise management level. Thus at the end of this paper, a systemic solution scheme, which employs the mode of "distributed computing - centralized storage - instant release", is brought forward, based on the combination of C/S and B/S release models. The architecture of the solution, the corresponding web technology and the client software are introduced. The application shows that the validity of this scheme.

The rapid assignment of ruminal fungi to presumptive genera using ITS1 and ITS2 RNA secondary structures to produce group-specific fingerprints

Danny S. Tuckwell, Matthew J. Nicholson, Christopher S. McSweeney, Michael K. Theodorou and Jayne L. Brookman (2005). The rapid assignment of ruminal fungi to presumptive genera using ITS1 and ITS2 RNA secondary structures to produce group-specific fingerprints. Microbiology, 151 (5) pp.1557-1567 Sponsorship: BBSRC / Stapledon Memorial Trust RAE2008

Discovering Clusters in Motion Time-Series Data

A new approach is proposed for clustering time-series data. The approach can be used to discover groupings of similar object motions that were observed in a video collection. A finite mixture of hidden Markov models (HMMs) is fitted to the motion data using the expectation-maximization (EM) framework. Previous approaches for HMM-based clustering employ a k-means formulation, where each sequence is assigned to only a single HMM. In contrast, the formulation presented in this paper allows each sequence to belong to more than a single HMM with some probability, and the hard decision about the sequence class membership can be deferred until a later time when such a decision is required. Experiments with simulated data demonstrate the benefit of using this EM-based approach when there is more "overlap" in the processes generating the data. Experiments with real data show the promising potential of HMM-based motion clustering in a number of applications.