A neuro-fuzzy system for sequence alignment on two levels

Soft Computing

Classification of classical melodic structures by style, composer, genre, period, etc., is a rather complex task. The level of difficulty varies across melodic frameworks. It would be interesting to see how we can impart this ability to a machine. This paper reports an improved pattern matching technique for composer and raga classification using a fuzzy analytical hierarchy process-based approach. The technique makes use of class-specific patterns extracted from a pattern discovery technique known as Structure Induction Algorithm for r superdiagonals and compactness trawler. Further, to represent inexact matches a modified matching technique is proposed to assign weights to the exact matching scores in a probabilistic manner. Subsequently, the weighted scores are fuzzified to quantify the extent of match. Finally, the fuzzy scores are aggregated and classified on the basis of minimum Euclidean distance from an ideal solution in the pattern space. Experiments conducted on datasets containing a wide range of melodies from classical western and classical Indian background show that the proposed technique exhibits consistently better classification success rate compared to the exact n-Gram-based approach and a widely used matching algorithm based on Levenshtein distance.

2010

We investigate the problem of matching symbolic representations directly to audio based representations for applications that use data from both domains. One such application is score alignment, which aligns a sequence of frames based on features such as chroma vectors and distance functions such as Euclidean distance. Good representations are critical, yet current systems use ad hoc constructions such as the chromagram that have been shown to work quite well. We investigate ways to learn chromagram-like representations that optimize the classification of “matching” vs. “non-matching” frame pairs of audio and MIDI. New representations learned automatically from examples not only perform better than the chromagram representation but they also reveal interesting projection structures that differ distinctly from the traditional chromagram.

2007

Melody is an important property for the perceptual description of Western musical pieces. In the monophonic context, retrieval systems based on melodic similarity generally consider sequences of pitches and durations. Algorithms that have been proposed for measuring melodic similarity rely on geometric representations, string matching techniques, etc. Adaptations of edit distance based algorithms, mainly applied in bioinformatic applications, to the musical domain have already been proposed. However, we present in this paper several experiments in order to optimize these methods. The different possible representations for pitches and durations are discussed and evaluated. Optimizations specific to musical applications are proposed and imply significant improvements of the algorithm. Evaluation of this algorithm led to the best results during the MIREX 2006 symbolic melodic similarity contest.

Journal of New Music Research, 2007

Melody is an important property for the perceptual description of Western musical pieces. In the monophonic context, retrieval systems based on melodic similarity generally consider sequences of pitches and durations. Algorithms that have been proposed for measuring melodic similarity rely on geometric representations, string matching techniques, etc. Adaptations of editing algorithms, mainly applied in bioinformatic applications, to the musical domain have already been proposed. However, we present in this paper several experiments in order to optimize these methods. The different possible representations for pitches and durations are discussed and evaluated. Optimizations specific to musical applications are proposed and imply significant improvements of the editing algorithm studied. Evaluation of this algorithm led to the best results during the MIREX 2006 symbolic melodic similarity contest.

2009

Existing symbolic music comparison systems generally consider monophonic music or monophonic reduction of polyphonic music. Adaptation of alignment algorithms to the music leads to accurate systems, but extensions to polyphonic music arise new problems. Indeed, a chord may match several notes, or the difference between two similar motifs may be a few swapped notes. Moreover, it is difficult to set up the substitution scores between chords. In this paper, we propose a general framework for polyphonic music which permits to directly apply the substitution score scheme set for monophonic music, and which allows new operations by extending the operations proposed by Mongeau and Sankoff [1]. From a practical point of view, the limitations of the size of chords and the number of notes that can be merged lead to a complexity that remains quadratic.

2008

Existing algorithms for estimating the music similarity generally consider monophonic music or reduced polyphonic music. Alignment algorithms are currently applied in different domains such as string matching or bioinformatics. Their applications to monophonic music leads to efficient and robust systems. In this paper, we propose to study and improve the adaptation of such alignment algorithms to polyphonic music, without information on the component voices of this music. A representation for polyphonic music, based on existing works that consider quotiented sequences, is proposed and justified with examples. Extensions for the alignment algorithm are also presented for taking into account the limiting problem of overlapped notes. Other limitations such as transposition invariance, multiple alignment problem are discussed. Experiments with symbolic performed music show the improvements induced by the adaptation proposed.

2007 IEEE Symposium on Computational Intelligence and Bioinformatics and Computational Biology, 2007

DNA matching is a crucial step in sequence alignment. Since sequence alignment is an approximate matching process there is a need for good approximate algorithms. The process of matching in sequence alignment is generally finding longest common subsequences. However, finding a longest common subsequence may not be the best solution for either a database match or an assembly. An optimal alignment of subsequences is based on several factors, such as quality of bases, length of overlap, etc. Factors such as quality indicate if the data is an actual read or an experimental error. Fuzzy logic allows tolerance of inexactness or errors in sub sequence matching. We propose fuzzy logic for approximate matching of subsequences. Fuzzy characteristic functions are derived for parameters that influence a match. We develop a prototype for a fuzzy assembler. The assembler is designed to work with low quality data which is generally rejected by most of the existing techniques. We test the assembler on sequences from two genome projects namely, Drosophila melanogaster and Arabidopsis thaliana. The results are compared with other assemblers. The fuzzy assembler successfully assembled sequences and performed similar and in some cases better than existing techniques.

2014

In the present work, the Fuzzy Inductive Reasoning methodology (FIR) is used to improve coherence among beat patterns, structured in a musical A-B form. Patterns were generated based on a probability matrix, encoding a particular musical style, designed by experts. Then, all possible patterns were generated and the most probables were selected. A-B musical forms were created and the coherence of the sequence was evaluated by experts by using linguistic quantities. The output pairs (A-B pattern and its qualification) were used as inputs to train a FIR system, and the variables that produce “coherent” outputs and the relations among them where identified as rules. The extracted rules are discussed in the context of the musical form and from the psychological perception.

This paper reports classification of classical Indian melodic structures (Ragas) using a Fuzzy Multi Attribute Decision Model constructed using string matching scores. Sixteen clippings belonging to four different Ragas were matched with standard templates of the Ragas with an aim to detect the occurrences of subsets of the standard templates within a test string. Matching score obtained with subsets of varying lengths (n-grams) have been analyzed using Fuzzy Analytical Hierarchy Process (FAHP). Matching scores were first fuzzified by assigning them fuzzy memberships. Subsequently, 2, 3 and 4 gram scores were chosen as criteria/attributes against which four alternatives were evaluated. This paper proposes the use of fuzzy entropy for calculating the relative weights for each fuzzy set in the FAHP model. It was observed that classification success rate improved significantly when the n-gram scores were fuzzified and the proposed technique was applied.

2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018

The alignment of heterogeneous sequential data (video to text) is an important and challenging problem. Standard techniques for this task, including Dynamic Time Warping (DTW) and Conditional Random Fields (CRFs), suffer from inherent drawbacks. Mainly, the Markov assumption implies that, given the immediate past, future alignment decisions are independent of further history. The separation between similarity computation and alignment decision also prevents end-to-end training. In this paper, we propose an end-to-end neural architecture where alignment actions are implemented as moving data between stacks of Long Short-term Memory (LSTM) blocks. This flexible architecture supports a large variety of alignment tasks, including one-to-one, one-to-many, skipping unmatched elements, and (with extensions) non-monotonic alignment. Extensive experiments on semi-synthetic and real datasets show that our algorithm outperforms state-of-the-art baselines. * The technique was conceived when all authors worked for Disney Research.