Publications
Discriminative N-gram selection for dialect recognition
September 6, 2009
Conference Paper
Published in:
INTERSPEECH 2009, 6-10 September 2009.
Topic:
R&D area:
Summary
Dialect recognition is a challenging and multifaceted problem. Distinguishing between dialects can rely upon many tiers of interpretation of speech data - e.g., prosodic, phonetic, spectral, and word. High-accuracy automatic methods for dialect recognition typically rely upon either phonetic or spectral characteristics of the input. A challenge with spectral system, such as those based on shifted-delta cepstral coefficients, is that they achieve good performance but do not provide insight into distinctive dialect features. In this work, a novel method based upon discriminative training and phone N- grams is proposed. This approach achieves excellent classification performance, fuses well with other systems, and has interpretable dialect characteristics in the phonetic tier. The method is demonstrated on data from the LDC and prior NIST language recognition evaluations. The method is also combined with spectral methods to demonstrate state-of-the-art performance in dialect recognition.
Summary
Dialect recognition is a challenging and multifaceted problem. Distinguishing between dialects can rely upon many tiers of interpretation of speech data - e.g., prosodic, phonetic, spectral, and word. High-accuracy automatic methods for dialect recognition typically rely upon either phonetic or spectral characteristics of the input. A challenge with spectral system...
READ MORE
The MIT Lincoln Laboratory 2008 speaker recognition system
September 6, 2009
Conference Paper
Published in:
INTERSPEECH 2009, 6-10 September 2009.
Topic:
R&D area:
Summary
In recent years methods for modeling and mitigating variational nuisances have been introduced and refined. A primary emphasis in this years NIST 2008 Speaker Recognition Evaluation (SRE) was to greatly expand the use of auxiliary microphones. This offered the additional channel variations which has been a historical challenge to speaker verification systems. In this paper we present the MIT Lincoln Laboratory Speaker Recognition system applied to the task in the NIST 2008 SRE. Our approach during the evaluation was two-fold: 1) Utilize recent advances in variational nuisance modeling (latent factor analysis and nuisance attribute projection) to allow our spectral speaker verification systems to better compensate for the channel variation introduced, and 2) fuse systems targeting the different linguistic tiers of information, high and low. The performance of the system is presented when applied on a NIST 2008 SRE task. Post evaluation analysis is conducted on the sub-task when interview microphones are present.
Summary
In recent years methods for modeling and mitigating variational nuisances have been introduced and refined. A primary emphasis in this years NIST 2008 Speaker Recognition Evaluation (SRE) was to greatly expand the use of auxiliary microphones. This offered the additional channel variations which has been a historical challenge to speaker...
READ MORE
Variability compensated support vector machines applied to speaker verification
September 6, 2009
Conference Paper
Published in:
INTERSPEECH 2009, Proc. of the 10th Annual Conf. of the Internatinoal Speech Communication Association, 6-9 September 2009, pp. 1555-1558.
Topic:
R&D area:
Summary
Speaker verification using SVMs has proven successful, specifically using the GSV Kernel [1] with nuisance attribute projection (NAP) [2]. Also, the recent popularity and success of joint factor analysis [3] has led to promising attempts to use speaker factors directly as SVM features [4]. NAP projection and the use of speaker factors with SVMs are methods of handling variability in SVM speaker verification: NAP by removing undesirable nuisance variability, and using the speaker factors by forcing the discrimination to be performed based on inter-speaker variability. These successes have led us to propose a new method we call variability compensated SVM (VCSVM) to handle both inter and intra-speaker variability directly in the SVM optimization. This is done by adding a regularized penalty to the optimization that biases the normal to the hyperplane to be orthogonal to the nuisance subspace or alternatively to the complement of the subspace containing the inter-speaker variability. This bias will attempt to ensure that inter-speaker variability is used in the recognition while intra-speaker variability is ignored. In this paper we present the theory and promising results on nuisance compensation.
Summary
Speaker verification using SVMs has proven successful, specifically using the GSV Kernel [1] with nuisance attribute projection (NAP) [2]. Also, the recent popularity and success of joint factor analysis [3] has led to promising attempts to use speaker factors directly as SVM features [4]. NAP projection and the use of...
READ MORE
Modeling and detection techniques for counter-terror social network analysis and intent recognition
March 7, 2009
Conference Paper
Published in:
2009 IEEE Aerospace Conf., March 7-14 2009.
Topic:
R&D area:
Summary
In this paper, we describe our approach and initial results on modeling, detection, and tracking of terrorist groups and their intents based on multimedia data. While research on automated information extraction from multimedia data has yielded significant progress in areas such as the extraction of entities, links, and events, less progress has been made in the development of automated tools for analyzing the results of information extraction to ?connect the dots.? Hence, our Counter-Terror Social Network Analysis and Intent Recognition (CT-SNAIR) work focuses on development of automated techniques and tools for detection and tracking of dynamically-changing terrorist networks as well as recognition of capability and potential intent. In addition to obtaining and working with real data for algorithm development and test, we have a major focus on modeling and simulation of terrorist attacks based on real information about past attacks. We describe the development and application of a new Terror Attack Description Language (TADL), which is used as a basis for modeling and simulation of terrorist attacks. Examples are shown which illustrate the use of TADL and a companion simulator based on a Hidden Markov Model (HMM) structure to generate transactions for attack scenarios drawn from real events. We also describe our techniques for generating realistic background clutter traffic to enable experiments to estimate performance in the presence of a mix of data. An important part of our effort is to produce scenarios and corpora for use in our own research, which can be shared with a community of researchers in this area. We describe our scenario and corpus development, including specific examples from the September 2004 bombing of the Australian embassy in Jakarta and a fictitious scenario which was developed in a prior project for research in social network analysis. The scenarios can be created by subject matter experts using a graphical editing tool. Given a set of time ordered transactions between actors, we employ social network analysis (SNA) algorithms as a filtering step to divide the actors into distinct communities before determining intent. This helps reduce clutter and enhances the ability to determine activities within a specific group. For modeling and simulation purposes, we generate random networks with structures and properties similar to real-world social networks. Modeling of background traffic is an important step in generating classifiers that can separate harmless activities from suspicious activity. An algorithm for recognition of simulated potential attack scenarios in clutter based on Support Vector Machine (SVM) techniques is presented. We show performance examples, including probability of detection versus probability of false alarm tradeoffs, for a range of system parameters.
Summary
In this paper, we describe our approach and initial results on modeling, detection, and tracking of terrorist groups and their intents based on multimedia data. While research on automated information extraction from multimedia data has yielded significant progress in areas such as the extraction of entities, links, and events, less...
READ MORE
Forensic speaker recognition: a need for caution
March 1, 2009
Journal Article
Published in:
IEEE Signal Process. Mag., Vol. 26, No. 2, March 2009, pp. 95-103.
Summary
There has long been a desire to be able to identify a person on the basis of his or her voice. For many years, judges, lawyers, detectives, and law enforcement agencies have wanted to use forensic voice authentication to investigate a suspect or to confirm a judgment of guilt or innocence. Challenges, realities, and cautions regarding the use of speaker recognition applied to forensic-quality samples are presented.
Summary
There has long been a desire to be able to identify a person on the basis of his or her voice. For many years, judges, lawyers, detectives, and law enforcement agencies have wanted to use forensic voice authentication to investigate a suspect or to confirm a judgment of guilt or...
READ MORE
Cognitive services for the user
January 1, 2009
Book Chapter
Published in:
Chapter 10, Cognitive Radio Technology, 2009, pp. 305-324.
Topic:
R&D area:
Summary
Software-defined cognitive radios (CRs) use voice as a primary input/output (I/O) modality and are expected to have substantial computational resources capable of supporting advanced speech- and audio-processing applications. This chapter extends previous work on speech applications (e.g., [1]) to cognitive services that enhance military mission capability by capitalizing on automatic processes, such as speech information extraction and understanding the environment. Such capabilities go beyond interaction with the intended user of the software-defined radio (SDR) - they extend to speech and audio applications that can be applied to information that has been extracted from voice and acoustic noise gathered from other users and entities in the environment. For example, in a military environment, situational awareness and understanding could be enhanced by informing users based on processing voice and noise from both friendly and hostile forces operating in a given battle space. This chapter provides a survey of a number of speech- and audio-processing technologies and their potential applications to CR, including: - A description of the technology and its current state of practice. - An explanation of how the technology is currently being applied, or could be applied, to CR. - Descriptions and concepts of operations for how the technology can be applied to benefit users of CRs. - A description of relevant future research directions for both the speech and audio technologies and their applications to CR. A pictorial overview of many of the core technologies with some applications presented in the following sections is shown in Figure 10.1. Also shown are some overlapping components between the technologies. For example, Gaussian mixture models (GMMs) and support vector machines (SVMs) are used in both speaker and language recognition technologies [2]. These technologies and components are described in further detail in the following sections. Speech and concierge cognitive services and their corresponding applications are covered in the following sections. The services covered include speaker recognition, language identification (LID), text-to-speech (TTS) conversion, speech-to-text (STT) conversion, machine translation (MT), background noise suppression, speech coding, speaker characterization, noise management, noise characterization, and concierge services. These technologies and their potential applications to CR are discussed at varying levels of detail commensurate with their innovation and utility.
Summary
Software-defined cognitive radios (CRs) use voice as a primary input/output (I/O) modality and are expected to have substantial computational resources capable of supporting advanced speech- and audio-processing applications. This chapter extends previous work on speech applications (e.g., [1]) to cognitive services that enhance military mission capability by capitalizing on automatic...
READ MORE
A comparison of subspace feature-domain methods for language recognition
September 22, 2008
Conference Paper
Published in:
INTERSPEECH 2008, 22-26 September 2008, pp. 309-312.
Topic:
R&D area:
Summary
Compensation of cepstral features for mismatch due to dissimilar train and test conditions has been critical for good performance in many speech applications. Mismatch is typically due to variability from changes in speaker, channel, gender, and environment. Common methods for compensation include RASTA, mean and variance normalization, VTLN, and feature warping. Recently, a new class of subspace methods for model compensation have become popular in language and speaker recognition--nuisance attribute projection (NAP) and factor analysis. A feature space version of latent factor analysis has been proposed. In this work, a feature space version of NAP is presented. This new approach, fNAP, is contrasted with feature domain latent factor analysis (fLFA). Both of these methods are applied to a NIST language recognition task. Results show the viability of the new fNAP method. Also, results indicate when the different methods perform best.
Summary
Compensation of cepstral features for mismatch due to dissimilar train and test conditions has been critical for good performance in many speech applications. Mismatch is typically due to variability from changes in speaker, channel, gender, and environment. Common methods for compensation include RASTA, mean and variance normalization, VTLN, and feature...
READ MORE
The MITLL NIST LRE 2007 language recognition system
September 22, 2008
Conference Paper
Published in:
INTERSPEECH 2008, 22-26 September 2008, 719-722.
Topic:
R&D area:
Summary
This paper presents a description of the MIT Lincoln Laboratory language recognition system submitted to the NIST 2007 Language Recognition Evaluation. This system consists of a fusion of four core recognizers, two based on tokenization and two based on spectral similarity. Results for NIST?s 14-language detection task are presented for both the closed-set and open-set tasks and for the 30, 10 and 3 second durations. On the 30 second 14-language closed set detection task, the system achieves a 1% equal error rate.
Summary
This paper presents a description of the MIT Lincoln Laboratory language recognition system submitted to the NIST 2007 Language Recognition Evaluation. This system consists of a fusion of four core recognizers, two based on tokenization and two based on spectral similarity. Results for NIST?s 14-language detection task are presented for...
READ MORE
A covariance kernel for SVM language recognition
March 31, 2008
Conference Paper
Published in:
ICASSP 2008, Proc. IEEE Int. Conf. on Acoustics, Speech and Signal Processing, 31 March - 4 April 2008, pp. 4141-4144.
Topic:
R&D area:
Summary
Discriminative training for language recognition has been a key tool for improving system performance. In addition, recognition directly from shifted-delta cepstral features has proven effective. A recent successful example of this paradigm is SVM-based discrimination of languages based on GMM mean supervectors (GSVs). GSVs are created through MAP adaptation of a universal background model (UBM) GMM. This work proposes a novel extension to this idea by extending the supervector framework to the covariances of the UBM. We demonstrate a new SVM kernel including this covariance structure. In addition, we propose a method for pushing SVM model parameters back to GMM models. These GMM models can be used as an alternate form of scoring. The new approach is demonstrated on a fourteen language task with substantial performance improvements over prior techniques.
Summary
Discriminative training for language recognition has been a key tool for improving system performance. In addition, recognition directly from shifted-delta cepstral features has proven effective. A recent successful example of this paradigm is SVM-based discrimination of languages based on GMM mean supervectors (GSVs). GSVs are created through MAP adaptation of...
READ MORE
A multi-class MLLR kernel for SVM speaker recognition
March 31, 2008
Conference Paper
Published in:
Proc. IEEE Int. Connf. on Acoustics, Speech and Signal Processing, ICASSP, 31 March - 4 April 2008, pp. 4117-4120.
Topic:
R&D area:
Summary
Speaker recognition using support vector machines (SVMs) with features derived from generative models has been shown to perform well. Typically, a universal background model (UBM) is adapted to each utterance yielding a set of features that are used in an SVM. We consider the case where the UBM is a Gaussian mixture model (GMM), and maximum likelihood linear regression (MLLR) adaptation is used to adapt the means of the UBM. Recent work has examined this setup for the case where a global MLLR transform is applied to all the mixture components of the GMM UBM. This work produced positive results that warrant examining this setup with multi-class MLLR adaptation, which groups the UBM mixture components into classes and applies a different transform to each class. This paper extends the MLLR/GMM framework to the multiclass case. Experiments on the NIST SRE 2006 corpus show that multi-class MLLR improves on global MLLR and that the proposed system?s performance is comparable with state of the art systems.
Summary
Speaker recognition using support vector machines (SVMs) with features derived from generative models has been shown to perform well. Typically, a universal background model (UBM) is adapted to each utterance yielding a set of features that are used in an SVM. We consider the case where the UBM is a...
READ MORE