FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W2265195717> ?p ?o ?g. }

• W2265195717 abstract "Abstract Speaker verification systems have shown significant progressand have reached a level of performance that make their usein practical applications possible. Nevertheless, large differ-ences in terms of performance are observed, depending on thespeaker or the speech excerpt used. This context emphasizesthe importance of a deeper analysis of the system’s performanceover average error rate. In this paper, the effect of the trainingexcerpt is investigated using ALIZE/SpkDet on two differentcorpora: NIST-SRE 08 (conversational speech) and BREF 120(controlled read speech). The results show that the SVS perfor-mance are highly dependent on the voice samples used to trainthe speaker model: the overall Equal Error Rate (EER) rangesfrom 4.1% to 29.1% on NIST-SRE 08 and from 1.0% to 33.0%on BREF 120. The hypothesis that such performance differ-ences are explained by phonetic contents of voice samples isstudied on BREF 120. 1. Introduction Over the last decade, automatic speaker verification systems(SVS) have been assessed regularly by the National Institute ofStandards and Technology (NIST) [1]. The evaluation focuseson text-independent speaker detection and offers a common ex-perimental protocol and a stable set of evaluation rules.Although the task difficulty is changing through years, the NISTcampaigns clearly show a drastic progress in terms of perfor-mance during the last years. The level of performance reachedby the systems has become suitable for a large set of practical,commercial applications. Many applications are already avail-able or planned for the next future, including the forensic ones.This context underlines the importance of a deep analysis ofthe system’s performance, for instance on a per speaker basis,while the performance is usually assessed only through averageerror rate. In addition to the average performance information,performance variability also needs to be evaluated. Indeed, theidentification of the performance variation factors is necessaryfor determining the contexts in which those systems may beused.The system performance is commonly measured using twokinds of errors. A false acceptance (FA) occurs when an impos-tor is accepted by the system. A false rejection (FR) consists ofrejecting a valid identity. Both error rates depend on the thresh-old used in the decision making process. Among the measuresused to compare system performances, detection error trade-off(DET) curve [2], Equal Error Rate (EER) and Decision CostFunction (DCF) are usually used. The DET curve is obtainedby plotting on a normal deviate curve the FA rate as a functionof the FR rate. The EER corresponds to the operating pointwhere FA rate = FR rate when the DCF corresponds to a spe-cific operating point, described by the weight tied to each error(FA and FR) and the prior probabilities of these errors. In NISTevaluation, each training excerpt is regarded as produced by adifferent speaker, but the same speaker may have been recordedin several extracts. No comparison between these different ex-tracts is conducted.Some studies have investigated the possible causes of perfor-mance variation. [3] showed that the performance of the systemmay be improved by increasing the length of the training andtesting signals. Indeed, the EER raised from 4.48% up to morethan 30% when the duration of the training signals are shortenedfrom 2.5 minutes to 10 seconds. Moreover, a short excerpt intraining is more disadvantageous than a short excerpt in testing(more than 14% of EER with short excerpts in testing vs. morethan 17% of EER with short excerpts in training)Inter-speaker variation has also been studied. Doddington et al.[4] studied the errors induced by different speakers in 12 auto-matic speaker verification systems, and showed that the topol-ogy of the errors depends on speakers, consistently from onesystem to another. They distinguished 4 types of speakers, il-lustrated by a ’menagerie’. Sheeps correspond to the defaultspeaker type (low FA, low FR). Goats are speakers who gen-erate a disproportionate false rejection rate. Lambs correspondto speakers who generate a disproportionate false alarm rate.Wolves correspond to speakers that are likely to be mistaken foran other speaker.Finally, the influence of the phonetic content of test excerptswas evaluated by [5]. Results suggest that glides and liquidstogether, vowels and more particularly nasal vowels and nasalconsonants contain more speaker-specific information than pho-netically balanced test utterances, even though the training ex-cerpt were composed of 15 seconds of phonetically balancedspeech.This paper focuses on the variability due to the signal sampleused to represent the speaker voice. The information about thespeaker may differ among training excerpts. The aim of this pa-per is to quantify the effect of such a variability on SVS perfor-mance. The SVS scores for each training excerpt are comparedin order to selected the best and the worst training excerpts.Global performance is assessed using two different databases.Section 2 describes the system used. Section 3 and 4 investi-gate on the effect of training excerpt on NIST-08 and Bref 120database respectively. A preliminary phonetic analysis on theBREF 120 database is conducted in section 5 before conclud-ing in section 6." @default.
• W2265195717 created "2016-06-24" @default.
• W2265195717 creator A5038496855 @default.
• W2265195717 creator A5047012047 @default.
• W2265195717 creator A5074739823 @default.
• W2265195717 creator A5080870411 @default.
• W2265195717 date "2010-06-28" @default.
• W2265195717 modified "2023-10-18" @default.
• W2265195717 title "Intra-speaker variability effects of Speaker Verification performance" @default.
• W2265195717 cites W1532958172 @default.
• W2265195717 cites W1533303231 @default.
• W2265195717 cites W158052051 @default.
• W2265195717 cites W1586405805 @default.
• W2265195717 cites W184563473 @default.
• W2265195717 cites W1863069154 @default.
• W2265195717 cites W2023828624 @default.
• W2265195717 cites W2045346852 @default.
• W2265195717 cites W2075933652 @default.
• W2265195717 cites W2395619215 @default.
• W2265195717 cites W2404548572 @default.
• W2265195717 cites W2404921859 @default.
• W2265195717 cites W2418001401 @default.
• W2265195717 hasPublicationYear "2010" @default.
• W2265195717 type Work @default.
• W2265195717 sameAs 2265195717 @default.
• W2265195717 citedByCount "6" @default.
• W2265195717 countsByYear W22651957172012 @default.
• W2265195717 countsByYear W22651957172014 @default.
• W2265195717 countsByYear W22651957172015 @default.
• W2265195717 countsByYear W22651957172016 @default.
• W2265195717 countsByYear W22651957172020 @default.
• W2265195717 crossrefType "proceedings-article" @default.
• W2265195717 hasAuthorship W2265195717A5038496855 @default.
• W2265195717 hasAuthorship W2265195717A5047012047 @default.
• W2265195717 hasAuthorship W2265195717A5074739823 @default.
• W2265195717 hasAuthorship W2265195717A5080870411 @default.
• W2265195717 hasConcept C133892786 @default.
• W2265195717 hasConcept C149838564 @default.
• W2265195717 hasConcept C28490314 @default.
• W2265195717 hasConcept C2982762665 @default.
• W2265195717 hasConcept C41008148 @default.
• W2265195717 hasConceptScore W2265195717C133892786 @default.
• W2265195717 hasConceptScore W2265195717C149838564 @default.
• W2265195717 hasConceptScore W2265195717C28490314 @default.
• W2265195717 hasConceptScore W2265195717C2982762665 @default.
• W2265195717 hasConceptScore W2265195717C41008148 @default.
• W2265195717 hasLocation W22651957171 @default.
• W2265195717 hasLocation W22651957172 @default.
• W2265195717 hasLocation W22651957173 @default.
• W2265195717 hasLocation W22651957174 @default.
• W2265195717 hasOpenAccess W2265195717 @default.
• W2265195717 hasPrimaryLocation W22651957171 @default.
• W2265195717 hasRelatedWork W1516392727 @default.
• W2265195717 hasRelatedWork W1521049138 @default.
• W2265195717 hasRelatedWork W2140022733 @default.
• W2265195717 hasRelatedWork W2150532155 @default.
• W2265195717 hasRelatedWork W2160753975 @default.
• W2265195717 hasRelatedWork W2253551762 @default.
• W2265195717 hasRelatedWork W2378903289 @default.
• W2265195717 hasRelatedWork W2911612049 @default.
• W2265195717 hasRelatedWork W3012250009 @default.
• W2265195717 hasRelatedWork W3048612421 @default.
• W2265195717 isParatext "false" @default.
• W2265195717 isRetracted "false" @default.
• W2265195717 magId "2265195717" @default.
• W2265195717 workType "article" @default.