Showing papers by "Ron Weiss published in 2010"

Model-Based Expectation-Maximization Source Separation and Localization

Abstract: This paper describes a system, referred to as model-based expectation-maximization source separation and localization (MESSL), for separating and localizing multiple sound sources from an underdetermined reverberant two-channel recording. By clustering individual spectrogram points based on their interaural phase and level differences, MESSL generates masks that can be used to isolate individual sound sources. We first describe a probabilistic model of interaural parameters that can be evaluated at individual spectrogram points. By creating a mixture of these models over sources and delays, the multi-source localization problem is reduced to a collection of single source problems. We derive an expectation-maximization algorithm for computing the maximum-likelihood parameters of this mixture model, and show that these parameters correspond well with interaural parameters measured in isolation. As a byproduct of fitting this mixture model, the algorithm creates probabilistic spectrogram masks that can be used for source separation. In simulated anechoic and reverberant environments, separations using MESSL produced on average a signal-to-distortion ratio 1.6 dB greater and perceptual evaluation of speech quality (PESQ) results 0.27 mean opinion score units greater than four comparable algorithms.

Identifying repeated patterns in music using sparse convolutive non-negative matrix factorization

Abstract: We describe an unsupervised, data-driven, method for auto- matically identifying repeated patterns in music by analyz- ing a feature matrix using a variant of sparse convolutive non-negative matrix factorization. We utilize sparsity con- straints to automatically identify the number of patterns and their lengths, parameters that would normally need to be fixed in advance. The proposed analysis is applied to beat- synchronous chromagrams in order to concurrently extract repeated harmonic motifs and their locations within a song. Finally, we show how this analysis can be used for long- term structure segmentation, resulting in an algorithm that is competitive with other state-of-the-art segmentation algo- rithms based on hidden Markov models and self similarity matrices. been on locating repetitions rather than on extracting of characteristic, repetitive patterns. Previous research on de- tecting motif occurrences across a collection (9) and cover- song retrieval based on short-snippets (3), illustrate the utility of extracting such patterns. In this paper we propose a novel approach for the auto- matic extraction and localization of repeated patterns in mu- sic audio. The approach is based on sparse shift-invariant probabilistic latent component analysis (14) (SI-PLCA), a probabilistic variant of convolutive non-negative matrix factorization (NMF). The algorithm treats a musical record- ing as a concatenation of a small subset of short, repeated patterns, and is able to simultaneously estimate both the patterns and their repetitions throughout the song. The anal- ysis naturally identifies the long-term harmonic structure within a song, while the short-term structure is encoded within the patterns themselves. Furthermore, we show how it is possible to utilize sparse prior distributions to learn the number of patterns and their respective lengths, min- imizing the number of parameters that must be specified exactly in advance. Finally, we explore the application of this approach to long-term segmentation of musical pieces. The remainder of this paper is organized as follows: Section 2 reviews the proposed analysis based on SI-PLCA and describes its relationship to NMF. Sections 3 and 4 describe prior distributions over the SI-PLCA parameters and the expectation maximization algorithm for parameter estimation. Sections 5 and 6 discuss how the proposed analysis can be used for structure segmentation and provide experimental results. Finally, we conclude in Section 7.

Exploring common variations in state of the art chord recognition systems

Abstract: Most automatic chord recognition systems follow a standard approach combining chroma feature extraction, filtering and pattern matching. However, despite much research, there is little understanding about the interaction between these different components, and the optimal parameterization of their variables. In this paper we perform a systematic evaluation including the most common variations in the literature. The goal is to gain insight into the potential and limitations of the standard approach, thus contributing to the identification of areas for future development in automatic chord recognition. In our study we find that filtering has a significant impact on performance, with self-transition penalties being the most important parameter; and that the benefits of using complex models are mostly, but not entirely, offset by an appropriate choice of filtering strategies.

Clustering beat-chroma patterns in a large music database

Abstract: A musical style or genre implies a set of common conventions and patterns combined and deployed in different ways to make individual musical pieces; for instance, most would agree that contemporary pop music is assembled from a relatively small palette of harmonic and melodic patterns. The purpose of this paper is to use a database of tens of thousands of songs in combination with a compact representation of melodic-harmonic content (the beatsynchronous chromagram) and data-mining tools (clustering) to attempt to explicitly catalog this palette ‐ at least within the limitations of the beat-chroma representation. We use online k-means clustering to summarize 3.7 million 4-beat bars in a codebook of a few hundred prototypes. By measuring how accurately such a quantized codebook can reconstruct the original data, we can quantify the degree of diversity (distortion as a function of codebook size) and temporal structure (i.e. the advantage gained by joint quantizing multiple frames) in this music. The most popular codewords themselves reveal the common chords used in the music. Finally, the quantized representation of music can be used for music retrieval tasks such as artist and genre classification, and identifying songs that are similar in terms of their melodic-harmonic content.

Synthetic biology: from modules to systems

Abstract: Synthetic biology is revolutionizing how we conceptualize and approach the engineering of biological systems. Recent advances in the field are allowing us to expand beyond the construction and analysis of small gene networks towards the implementation of complex multicellular systems with a variety of applications. In this talk I will describe our integrated computational / experimental approach to engineering complex behavior in living systems ranging from bacteria to stem cells. In our research, we appropriate useful design principles from electrical engineering and other well established fields. These principles include abstraction, standardization, modularity, and computer aided design. But we also spend considerable effort towards understanding what makes synthetic biology different from all other existing engineering disciplines and discovering new design and construction rules that are effective for this unique discipline.We will briefly describe the implementation of genetic circuits with finely-tuned digital and analog behavior and the use of artificial cell-cell communication to coordinate the behavior of cell populations for programmed pattern formation. Recent results with implementing Turing patterns with engineering bacteria will be presented. Arguably the most significant contribution of synthetic biology will be in medical applications such as tissue engineering. We will discuss preliminary experimental results for obtaining precise spatiotemporal control over stem cell differentiation. For this purpose, we couple elements for gene regulation, cell fate determination, signal processing, and artificial cell-cell communication. We will conclude by discussing the design and preliminary results for creating an artificial tissue homeostasis system where genetically engineered stem cells maintain indefinitely a desired level of pancreatic beta cells despite attacks by the autoimmune response. The system, which relies on artificial cell-cell communication, various regulatory network motifs, and programmed differentiation into beta cells, may one day be useful for the treatment (or cure) of diabetes.

Bio-field programmable gate array and bio-programmable logic array: reconfigurable chassis construction

Abstract: Aspects of the invention relate to reconfigurable chassis that allow for rapid construction and optimization of biocircuits.

Gordon Music Collection Database Management System

Abstract: We present Gordon, a multi-platform database tool for managing very large music collections, targeted toward music information retrieval research. Gordon contains a unified framework for managing audio files in any format, matched with corresponding artist, album, and track metadata which can be resolved against the MusicBrainz service. Also supported are per-track annotations, useful for storing e.g. reference chord transcriptions alongside the corresponding audio. Currently in development is a framework for feature extraction and caching.

An RNAi-Enhanced Logic Circuit for Cancer Specific Detection and Destruction

Abstract: : Modern breast cancer therapies utilize non-specific approaches to kill or remove cancerous cells, inflicting significant collateral damage to healthy cells. In response to the need for highly targeted detection and destruction of cancerous cells, we propose to implement multi-input genetic circuits that act as cell state classifiers based on mRNA or microRNA expression profiling. The mRNA sensing project is focused on the MCF-7 breast adenocarcinoma cell line. MCF-7 cells overexpress Gata3, NPY1R and TFF1 mRNA relative to healthy cells. Based on our bioinformatics analysis, taking into account the three biomarkers allows for dramatically improved specificity in comparison to targeting single genes. We therefore design a three-input AND gate that triggers a response only when all three biomarkers are expressed above a defined threshold. In second approach we implement transcriptional/post-transcriptional regulatory circuit that senses expression levels of a customizable set of endogenous microRNAs and computes whether to trigger a response if the expression levels match a pre-determined profile of interest. We have created a circuit that computes a complex abstract logic (miR1 AND miR2+3 AND NOT miR4 AND NOTmiR5 AND NOT miR6) and selectively triggers output response in HeLa but not in other cells.