This RFC is not a candidate for any level of Internet Standard.The IETF disclaims any knowledge of the fitness of this RFC for any purpose and in particular notes that the decision to publish is not based on IETF review for such things as security, congestion control, or inappropriate interaction with deployed protocols.The

We have developed a system that allows us to index and deliver audio and video over the Internet. The system has been in continuous operation since March 1998 within the company. The design of our system differs from previous systems because 1) the indexing can be based on an annotation stream generated by robust transcript alignment, as well as closed captions, and 2) it is a distributed system that is designed for scalable, high performance, universal access through the World Wide Web. Extensive tests of the system show that it achieves a performance level required for Internet-wide delivery. This paper discusses our approach to the problem, the design requirements, the system architecture, and performance figures. It concludes by showing how the next generation of annotations from speech recognition and computer vision can be incorporated into the system.

While text indexing on the Web and corporate intranets has received considerable attention, indexing of multimedia documents in the same environments is still in its nascent stages. We have developed a system that allows us to index and serve multimedia documents over the Internet. The system has been in continuous operation since March 1998 within the company and has been tested extensively under heavy load conditions. In this paper we describe the design and implementation of the system and we present a user-model study indicating the usage patterns that we encountered during the system deployment. The main part of the paper presents the results of a detailed study that sheds light into the performance characteristics of the system components, identifies bottlenecks, and provides information to systems designers interested in delivering systems with good performance for this important class of applications. We find that the limiting factor for the three major subcomponents that comprise the indexing service is the memory system performance, while the component responsible for delivering the video/audio bits to the users is only limited by the bandwidth of our Internet connection. Surprisingly, I/O is not a bottleneck on our system despite the large amount of disk space consumed by the multimedia data. c Compaq Computer Corporation, 1999 This work may not be copied or reproduced in whole or in part for any commercial purpose. Permission to copy in whole or in part without payment of fee is granted for nonprofit educational and research purposes provided that all such whole or partial copies include the following: a notice that such copying is by permission of the Cambridge Research Laboratory of Compaq Computer Corporation in Cambridge, Massachusetts; an acknowledgment of the authors and individual contributors to the work; and all applicable portions of the copyright notice. Copying, reproducing, or republishing for any other purpose shall require a license with payment of fee to the Cambridge Research Laboratory. All rights reserved. CRL Technical reports are available on the CRL’s web page at http://www.crl.research.digital.com. Compaq Computer Corporation Cambridge Research Laboratory One Kendall Square, Building 700 Cambridge, Massachusetts 02139 USA

A New Algorithm for Robust Sp eech Recognition: The Delta VectorTaylor Series ApproachPedroJ.Moreno and Brian Ebermanemail: pjm@crl.dec.com, bse@crl.deomDigital Equipment Corp orationCambridge Research Lab oratoryABSTRACTIn this pap er we present a new mo del-based comp ensationtechnique called Delta Vector Taylor Series (DVTS). Thisnew technique is an extension and improvementoer theVector Taylor Series (VTS) approach [7] that addressesseveral of its limitations .In particular, we presentanew statistical representation for the distribution of cleansp eech feature vectors based on a weighted vector co de-b o ok. This change to the underlying probabili ty densityfunction (PDF) allows us to pro duce more accurate andstable solutions for our algorithm. The algorithm is alsopresented in a EM-MAP framework where some the en-vironmental parameters are treated as random variableswith known PDF's. Finally,we explore a new comp ensa-tion approach based on the use of convex hulls.Weevaluate our algorithm in a phonetic classi cati on taskon the TIMIT [5] database and also in a small vo cabu-lary size sp eech recognition database. In b oth databasesarti cial and natural noise is injected at several signal tonoise ratios (SNR). The algorithm achieves matched p er-formance at all SNR's ab ove 10 dB.1.Intro ductionOver the last years several techniques have b een prop osedto deal with the problem of sp eech recognition in noisy en-vironments. Some of them such as PMC [3], or MLLR [6]have used the recognition engine and its rich statisticalrepresentation (more than 90,000 Gaussians in systemslike SPHINX-3 and HTK [9]) to mo del and comp ensatefor the e ects of the environment on sp eech recognitionsystems. Other techniques like CDCN [1] and POF [8]among others have used a reduced set of Gaussian mix-tures (typically 256 or less) to mo del the clean sp eechfeature vectors and prepro cess the noisy sp eech featuresvectors to e ectively "clean" the features b efore b eing pro-cessed by the recognition engine.The use of a rich statistical representation improves p er-formance, but has the drawback of using the whole sp eechrecognition engine with its asso ciated complexity.Anideal robust recognition technique should have the advan-tages of a rich statistical representation and at the sametime b eing simple and fast in its op eration.The Delta Vector Taylor Series (DVTS) approachis anattempt in this direction.It tries to gain the b ene tsof a rich statistical representation and a low complexitytechnique for robust sp eech recognition. It tries to achievethese goals by using a di erent statistical representationfor the sp eech feature vectors.The outline of the pap er is as follows. In section 2 wedescrib e the DVTS algorithm.In section 3 we brieydescrib e the necessary mo di cations to the algorithm tomakeit work as a lter.In section 4 we describ e ourexp erimental results and nally in section 5 we presentour conclusions.2.New Algorithm: Delta-VTSDVTS mo dels the sp eech feature vectors as a weightedsum of multidimensi onal Dirac deltasp(x)=M1Xk=0P[k])(1)where eachvector function(xk) is mo deled as(xk)=D1Yi=0ii;k)(2)P[k]is ana prioriprobability of observing a particulardelta. The sum of these probabili ties must add up to one.This novel representation of the PDF ofxhas several ad-vantages. First of all it greatly simpli es the mathemat-ical assumptions of the VTS [7] algorithm. It pro ducesa simple, fast, robust and direct formulation of the EMsolutions already presented in [7].In this pap er we assume a mo del of the environmentinwhich sp eech is corrupted by unknown additive stationarynoise and unknown linear lteringZ(!)=X)jH2+N(3)whereZ(!) represents the p ower sp ectrum of the de-graded sp eech,X(!) is the p ower sp ectrum of the cleansp eech,jH(!)2is the transfer function of the linear lter,andN(!) is the p ower sp ectrum of the additive noise.In the log-mel-sp ectral domain this can b e expressed asz=x+ log (exp (q) + exp (n))(4)or in more general termsz=x+f(;nq)(5)

The performance of speaker recognition algorithms drops signi(cid:12)cantly when testing and training acoustic environments di(cid:11)er. This decrease is caused by the statistical mismatch between the statistics representing the speaker and the testing acoustic data. This paper reports our preliminary results on the application of a novel environmental compensation algorithm to the problem of speaker recognition and identi(cid:12)cation. This new technique, called the Delta Vector Taylor Series (DVTS) approach, improves performance at signal-to-noise ratios below 20dB. The algorithm imposes a model of how the environment modi(cid:12)es speaker statistics and uses Expectation-Maximization (EM) to solve a joint maximum likelihood formulation for the speaker recognition problem over both the speakers and the environment. We report experi-mental results on a subset of the TIMIT and NTIMIT database.