Introduction This OSAC Speaker Recognition Process Map .
OSAC Speaker RecognitionSubcommitteeProcess Map of Current Practices inForensic Speaker RecognitionSeptember 30, 2019 11:22 AMPage 1 of 32IntroductionThis OSAC Speaker Recognition process map arose from the need to establish a common frame ofreference to help overcome differences in terminology and participants' background andexperience. The development of the process map helped the participants to better understandcurrent practices and communicate them in a constructive way.Representatives of multiple U.S. government agencies, individual practitioners, and internationalexperts met for three days with a facilitator to create the first draft that sketched the componentsof a forensic examination. The current version incorporates additional contributions from a varietyof researchers and practitioners. The OSAC Speaker Recognition subcommittee would like toacknowledge and thank all those who participated in the development of this process map.The process depicted does not represent the practice of any single laboratory, but generalizes thediverse practices of multiple laboratories. This document reflects a balance between an attempt tobe comprehensive and the efficient use of volunteers' time. It is intended to be descriptive only,and its release does not imply endorsement by the OSAC Speaker Recognition Subcommittee of anyspecific approach or process. No inferences should be drawn from the inclusion or exclusion of anyapproach or process or from the level of detail provided for any particular approach or process.This process map is not intended to represent a best practice but rather to facilitate thedevelopment of future best practice documents by the OSAC Speaker Recognition Subcommittee.This document is a work product of the OSAC Speaker Recognition Subcommittee.
OSAC Speaker Recognition SubcommitteeInquiry ReceivedProcess Map of Current Practices in Forensic Speaker Recognition1200CaseAcceptance1100Case ember 30, 2019 11:22 AM Page 2 of 323100Pre-AnalysisObservationsand cMethod4300Holistic xpert-DrivenAuditory Phonetic& AcousticPhonetic ctrographic5200Verification5300Case Close-OutTerminate CaseCommentary4600Blind GroupingLegendCommentaryProcess start/endMultistep subprocessSelection from multiple optionsThis document is a work product of the OSAC Speaker Recognition Subcommittee.
OSAC Speaker Recognition SubcommitteeProcess Map of Current Practices in Forensic Speaker Recognition )September 30, 2019 11:22 AMPage 3 of 32Return to OverviewLegendProcess start/endSingle process stepInquiry Received Approval received?Analysis timeframe appropriate?Technically feasible?Sufficient quality?Sufficient quantity?Knowns collected properly?Analysis requested appropriate?Resources available to address bias?Multistep process that may be pre-defined in astandard, by lab policy, and/or by examiners1120All case acceptancecriteria met?Y1220Assign CasePriorityIndicates that the next or previous step issomewhere else on the process mapNTo5300Commentary1210RequestReceived?Decision stepY1230Log Chain ofCustody1240Will foils (imposters)be used in thisexamination?Y1250Are there appropriatefoils available in thiscase?Y1260Use foils fromcase1270Assign Case1280Will case be worked/assessed by more thanone examiner?NNTo5300N1255Obtain foilsamples(per agencypolicy)N1285Transfer Case toAssigned ExaminerCommentaryThis document is a work product of the OSAC Speaker Recognition Subcommittee.Y1290Transfer Case to AssignedExaminer ensuring that noinformation about the casehas been given tosecondary analystTo2000
OSAC Speaker Recognition SubcommitteeProcess Map of Current Practices in Forensic Speaker RecognitionSeptember 30, 2019 11:22 AM Page 4 of 32Return to OverviewLegendSingle process stepMultistep process that may be pre-defined in astandard, by lab policy, and/or by examinersFrom10002110Determine PhysicalControls and DataHandling Requirements(Gloves, virus checks,write protection, etc)Take steps to protect theevidence (make aworking copy)Decision step2130Optimizing Playback(finding the best wayto playback media)2120Inventory(mark and describeevidence per agencypolicy)Indicates that the next or previous step issomewhere else on the process map(see SWGDE BestPractices for ForensicAudio)Commentary2210Best AvailableData?2240Expert ConsultNeeded?(translator, videoanalyst, etc)Y2250Consult ExpertNN2220Communicate withSubmitter or checkarchives to obtain bettersamplesY2230Can I proceed withcurrent or obtainedsamples?2260Can I continue withexamination?NYNTo5300CommentaryThis document is a work product of the OSAC Speaker Recognition Subcommittee.YTo3000
OSAC Speaker Recognition SubcommitteeProcess Map of Current Practices in Forensic Speaker Recognition )September 30, 2019 11:22 AMPage 5 of 32Return to OverviewFrom20003105Select question ofinterest for analysis3110Observe intrinsic andextrinsic properties inQ and K(Q is the limitingevidence)From51003115Document intrinsicand extrinsicproperties in Q and K3120Do speaking styleand recordingconditions matchbetween Qand K?N3125Document thematching andmismatchingconditions andfeatures of the Qand K, as applicableY3130YDo I have amethod that workswith these data?3145Select appropriatemethod(s)3150Will processingimproveperformance ofmethod?3155Process Data(see list ofprocessingtechniques)Y3160Did processingelucidate any newintrinsic or extrinsicproperties?3135Will processingpermit a methodselection?Y3165Document newlyobserved 3210Do I haveappropriate data tocarry out downstreamprocessing?3250Continue to method(consider contextual biaswhen deciding whichmethod to use first)YN3220Can I obtain orsimulate comparabletest data?LegendSingle process stepMultistep process that may be pre-defined in astandard, by lab policy, and/or by examinersN3240Obtain orsimulaterequired dataY3230Can I continue withexamination?YNDecision stepTo5300Process step that results in documentationCommentaryIndicates that the next or previous step issomewhere else on the process mapThis document is a work product of the OSAC Speaker Recognition Subcommittee.To4000
OSAC Speaker Recognition SubcommitteeProcess Map of Current Practices in Forensic Speaker RecognitionSeptember 30, 2019 11:22 AM Page 6 of 32Return to OverviewAcoustic Phonetic Statistical Analysis (Semiautomatic Analysis)4110Assess suitability ofHuman SupervisedAutomatic Analysis4115Is this methodappropriate?YN4120Assess suitability ofHolistic AuditoryPerceptual Analysis4125Is this methodappropriate?To4100YNFrom30004130Assess suitability ofExpert-Driven AuditoryPhonetic and AcousticPhonetic Analysis4135Is this methodappropriate?4145Is this methodappropriate?Y4155Is this ess suitability of BlindGrouping MethodTo4300To4100N4140Assess suitability ofSpectrographic AnalysisTo4200Acoustic Phonetic Statistical Analysis (or Semiautomatic Analysis)is similar to Human Supervised Automatic Analysis (4200), butuses features derived via phonetic analysis, including humansupervised measurements of acoustic properties of the gle process stepDecision stepIndicates that the next or previous stepis somewhere else on the process mapThis document is a work product of the OSAC Speaker Recognition Subcommittee.
OSAC Speaker Recognition SubcommitteeProcess Map of Current Practices in Forensic Speaker Recognition )September 30, 2019 11:22 AMPage 7 of 32Return to Overview 4000NFrom41004205Is diarizationneeded?4210Diarize speakersin recordingY4215Optimize theselection4220Create new filecontaining onespeaker perrecording4223Have I separatedall speakers ofinterest?4225Can I continue withexamination?(data sufficiency)YY4230Conduct expertcritical listening(set expectationsand parameterselection)NN4235Do I have a systemoptimized for mycase data?Y4237Select ASR(based on classifier/algorithm,hyperparameter/model, referencepopulation, settings)4238Train system usingtraining dataYN4245Perform validationtest?Y4241Train systemusing trainingdata4248Performanceadequate?YN4240Can I optimize myautomated speakerrecognition systemfor the caseconditions?4247Reportvalidationresults4246Test systemusing validationdata4242Test N4244Make changesto systemN4275Report that systemperformance under the caseconditions is inadequate toproceed with an evaluation ofthe strength of evidenceassociated with the knownand questioned-speakerrecordingsN4249Apply system toknown-speaker sults?Y4280Report onlynumeric result (e.g.score, likelihoodratio)?N4255Interpret ASR output(refer to distributionson comparablesystems and data andassociated error rates)NY4282Convert resultto verbaldescriptionN4270Adjust model/referencepopulation and/orsettings.N4265Stop analysis?4290Complete HumanSupervised AutomaticMethod? (formerlyRun data on anotherASR?)N4260Is the outputlogically consistentwith regards to typeof mismatch?4257Assessconfidence inASR resultsY4285Report resultsYCommentaryThis document is a work product of the OSAC Speaker Recognition Subcommittee.4267Document issues notedin box 4260YTo5000
OSAC Speaker Recognition SubcommitteeReturn to OverviewFrom4100Process Map of Current Practices in Forensic Speaker RecognitionSeptember 30, 2019 11:22 AM Page 8 of 3240004310Is diarizationneeded?Y4320Diarize speakersin recording4330Optimize theselection4340Create new filecontaining one speakerper recording4350Have I segmentedall speakers ofinterest?YNN4360Can I continuewith examination?(sufficient data)Y4370Conduct expertcritical listening(set expectationsand parameterselection)4380Document resultsTo5000NCommentaryLegendSingle process stepMultistep process that may be pre-defined ina standard, by lab policy, and/or by examinersDecision stepProcess step that results in documentationIndicates that the next or previous step issomewhere else on the process mapThis document is a work product of the OSAC Speaker Recognition Subcommittee.
OSAC Speaker Recognition SubcommitteeProcess Map of Current Practices in Forensic Speaker Recognition )September 30, 2019 11:22 AMPage 9 of 32Return to Overview 4000From41004405Do I have thelanguage, phonetic,sociolinguistic, dialectexpertise appropriatefor data?Y4410Is diarizationneeded?Y4412Diarize speakersin recording4415Optimize theselection4420Create new filecontaining onespeaker perrecording4425Have I segmentedall speakers ofinterest?NYNN4430Can I continue withexamination?(sufficient data)NY4435Conduct expertcritical listening(set expectationsand rements offeatures y/Extractauditory/acousticfeatures of interest4450Use automationto evaluatetypicality?NY4452Selectappropriate toolto tetypicalityN4465Use automationto evaluatesimilarity?Y4467Selectappropriate toolto ntsimilarity4480Is the outputlogicallyconsistent ?N4490Document resultsYNN4485Is analysiscomplete?(formerly Rerun/select anothertool?)YTo5000CommentaryLegendSingle process stepMultistep process that may be pre-defined in astandard, by lab policy, and/or by examinersDecision stepProcess step that results in documentationIndicates that the next or previous step issomewhere else on the process mapThis document is a work product of the OSAC Speaker Recognition Subcommittee.
OSAC Speaker Recognition SubcommitteeProcess Map of Current Practices in Forensic Speaker RecognitionSeptember 30, 2019 11:22 AM Page 10 of 32Return to Overview 4000From41004510Conduct examinationaccording to standard(e.g. IAI Standard,reference JFI (1991) 41:5)4520Document resultsTo5000CommentaryLegendSingle process stepMultistep process that may be pre-defined in astandard, by lab policy, and/or by examinersIndicates that the next or previous step issomewhere else on the process mapThis document is a work product of the OSAC Speaker Recognition Subcommittee.
OSAC Speaker Recognition SubcommitteeProcess Map of Current Practices in Forensic Speaker Recognition )September 30, 2019 11:22 AMPage 11 of 32Return to Overview 4000From41004605Are there appropriatefoils available in thecase?4610Create a list of speakers tobe included in the blindgrouping, containing atleast each Q, and K and 0or more foils.Y4615Is diarizationneeded?Y4620Diarize speakersin recordings4630Create new file per speakerper recording. Strip therecording from any speechthat contains name,addresses, etc. that wouldhelp linking fragments.4625Optimize theselectionN4635Is it possible to make 1 ormore fragments of max 20seconds for each speakerand each recording?NYN4640Is it possible to maketrue same speaker and truedifferent speakerpairs?Y4645Create a list of fragmentsper speaker per recording,max 20 fragments.4650Create the max 20 fragments intoseparate audio files, all with thesame audio specs (take the onewith the highest quality).N4670Compare each of the fragmentswith each other and group themby perceived speaker identity.4655Is there enough speakervariation in each of thefragments (so as not to biasby selection)?Y4660Anonymize and collect all thefragments in one larger audio filein random order (but make alookup table to retrieve theorigins). Present the larger file toresearcher #2.4665Document compositionof larger audio file.N4675Is each of the fragmentssuitable for grouping?N4685Document a grouping,including perceivedsimilarity within groupsand perceived dissimilaritybetween groups and theungrouped fragments.4680Separate out the unsuitablefragments.4690Interpret grouping usinglookup table and includingcorrectness of grouping inground truth.4695Document resultsTo5000YCommentaryLegendSingle process stepMultistep process that may be pre-defined ina standard, by lab policy, and/or by examinersDecision stepProcess step that results in documentationIndicates that the next or previous step issomewhere else on the process mapThis document is a work product of the OSAC Speaker Recognition Subcommittee.
OSAC Speaker Recognition SubcommitteeProcess Map of Current Practices in Forensic Speaker Recognition )September 30, 2019 11:22 AMPage 12 of 32Return to sconfidence foreach methodused5115YCombine (fuse)results?5120Assign weightand combine(fuse) results peragency policy5150YWillconsultation behelpful?5125Formulateconclusion5155Share casedetailSeek input5160Will additionalanalysis behelpful?NNNNY5165Document Opinionor ConclusionsTo40005170Are all casequestionsanswered?Y5180Draft preliminaryreportNTo3000Commentary5205Verify conclusions?(according to agencypolicy)Y5210Conduct blindreview?Y5220Conduct is(examinerbegins at 310)5250Consensusconclusionreached?Y5270Does the reportneed to bechanged?Y5280Revise Reportbased onconsensus review5290Conductadministrativereview5310Notify requestor andtransmit report,where appropriateTerminate ReviewNN5225Share limited casedata (per agencypolicy)5230Conducttechnicalreview (peragency policy)5260Resolve conflicts(Follow agencyconflictresolution policy )CommentaryLegendCommentarySingle process stepMultistep process that may be pre-defined ina standard, by lab policy, and/or by examinersDecision stepProcess step that results in documentationIndicates that the next or previous step issomewhere else on the process mapThis document is a work product of the OSAC Speaker Recognition Subcommittee.
OSAC Speaker RecognitionSubcommitteeProcess Map of Current Practices inForensic Speaker RecognitionSeptember 30, 2019 11:22 AMPage 13 of 32Return to Overview 1000Process Step1000 – Administrative AssessmentDescriptionTerms and r 30, 2019 11:22 AMThis document is a work product of the OSAC Speaker Recognition Subcommittee.
OSAC Speaker RecognitionSubcommitteeProcess Map of Current Practices inForensic Speaker RecognitionSeptember 30, 2019 11:22 AMPage 14 of 32Return to Overview 1000 1100Process Step1100 – Case SuitabilityDescriptionTerms and r 30, 2019 11:22 AMThis document is a work product of the OSAC Speaker Recognition Subcommittee.
OSAC Speaker RecognitionSubcommitteeProcess Map of Current Practices inForensic Speaker RecognitionSeptember 30, 2019 11:22 AMPage 15 of 32Return to Overview 1000 1200Process Step1200 – Case AcceptanceDescriptionTerms and r 30, 2019 11:22 AMThis document is a work product of the OSAC Speaker Recognition Subcommittee.
OSAC Speaker RecognitionSubcommitteeProcess Map of Current Practices inForensic Speaker RecognitionSeptember 30, 2019 11:22 AMPage 16 of 32Return to Overview 2000Process Step2000 – Technical AssessmentDescriptionTerms and r 30, 2019 11:22 AMThis document is a work product of the OSAC Speaker Recognition Subcommittee.
OSAC Speaker RecognitionSubcommitteeProcess Map of Current Practices inForensic Speaker RecognitionSeptember 30, 2019 11:22 AMPage 17 of 32Return to Overview 2000 2100Process Step2100 – Preliminary EvaluationDescriptionTerms and DefinitionsCommentsIssuesReferencesSWGDE Best Practices for Forensic Audio.RevisedSeptember 30, 2019 11:22 AMThis document is a work product of the OSAC Speaker Recognition Subcommittee.
OSAC Speaker RecognitionSubcommitteeProcess Map of Current Practices inForensic Speaker RecognitionSeptember 30, 2019 11:22 AMPage 18 of 32Return to Overview 2000 2200Process Step2200 – Content ReviewDescriptionTerms and r 30, 2019 11:22 AMThis document is a work product of the OSAC Speaker Recognition Subcommittee.
OSAC Speaker RecognitionSubcommitteeProcess Map of Current Practices inForensic Speaker RecognitionSeptember 30, 2019 11:22 AMPage 19 of 32Return to Overview 3000Process Step3000 – ProcessingDescriptionTerms and r 30, 2019 11:22 AMThis document is a work product of the OSAC Speaker Recognition Subcommittee.
OSAC Speaker RecognitionSubcommitteeProcess Map of Current Practices inForensic Speaker RecognitionSeptember 30, 2019 11:22 AMPage 20 of 32Return to Overview 3000 3100Process Step3100 – Pre-Analysis Observations and ProcessingDescriptionTerms and Definitions3155: Types of processing Enhancement for intelligibility or listenability (e.g. tone removal, spectral shaping, adaptive filtering, etc.) Normalization Convert sampling rate / bit depth Channel conversion DC offset ber 30, 2019 11:22 AMThis document is a work product of the OSAC Speaker Recognition Subcommittee.
OSAC Speaker RecognitionSubcommitteeProcess Map of Current Practices inForensic Speaker RecognitionSeptember 30, 2019 11:22 AMPage 21 of 32Return to Overview 3000 3200Process Step3200 – Relevant Population DataDescriptionThis block describes the necessary use of different data sets for testing system performance . Evaluation typically requires training data andtest data sets, but other data may be required (e.g. for calibration).Term ands DefinitionsCommentsIssuesReferencesMorrison, G.S., Thompson, W.C. (2017). Assessing the admissibility of a new generation of forensic voice comparison testimony. ColumbiaScience and Technology Law Review, 18, 326–434 §3.1. http://www.stlr.org/cite.cgi?volume 18&article morrisonThompson(Preprints: https://ssrn.com/abstract 2883767 https://www.newton.ac.uk/files/prep
4200 Human Supervised Automatic Method 4300 Holistic Auditory Perceptual 4500 Spectrographic 4400 Expert-Driven Auditory Phonetic & Acoustic Phonetic Analysis 5100 Evaluation/ Generating Conclusion 5200 Verification 5300 Case Close-Out Terminate Case Commentary Commentary Commentary Commentary Commentary 4
to answers A–F. There is one extra answer. Speaker 1 Speaker 2 Speaker 3 Speaker 4 Speaker 5 A The speaker is inspired by Jessica. B The speaker is critical of Jessica’s parents. C The speaker congratulates Jessica. D The speaker describes the event. E The speaker comments on how Jessica looks. F The speaker knows Jessica personally.
† [7] (LYNXR-EN) that LYNX has finished For LYNXR/LYNXR24 only options 0, 1, 2, and 3 are applicable Central Dialing Mode Station Pulse Tone Pulse Tone No WATS 0 No Speaker Phone 1 No Speaker Phone 4 With Speaker Phone 5 With Speaker Phone WATS 2 No Speaker Phone 3 No Speaker Phone 6 With Speaker Phone 7 With Speaker Phone 48 REPORT FORMAT for PRIM./SEC [7, 7] Primary .
13. Documentation, Case Record and Report Contents 14. Ensuring that Forensic Analysis is Based Upon Task-Relevant Information 15. Forensic Science Curriculum Development . (ENFSI) might interface with OSAC efforts. ENFSI has 64 member institutes, 2 standing committees, 17 working groups, and a 2
ow and speech recognition toolkit Kaldi. After reproducing state-of-the-art speech and speaker recognition performance using TIK, I then developed a uni ed model, JointDNN, that is trained jointly for speech and speaker recognition. Experimental results show that the joint model can e ectively perform ASR and SRE tasks. In particular, ex-
The Ultimate Guide to Employee Rewards & Recognition v1.0. Table of contents INTRODUCTION 3 EVOLVING ROLE OF HR 5 REWARDS VS RECOGNITION 8 BENEFITS OF REWARDS AND RECOGNITION 10 TECHNOLOGY IN REWARDS AND RECOGNITION 15 A CULTURE OF PEER TO PEER RECOGNITION 26 SELECTING A REWARDS AND RECOGNITION PLATFORM 30
the House of the people 94. Vacation and resignation of, and removal from, the offices of speaker and deputy speaker 95. Power of the deputy speaker or other person to perform the duties of the office of, or to act as, speaker 96. The speaker or the deputy speaker not to preside while a resolution for
18-794 Pattern Recognition Theory! Speech recognition! Optical character recognition (OCR)! Fingerprint recognition! Face recognition! Automatic target recognition! Biomedical image analysis Objective: To provide the background and techniques needed for pattern classification For advanced UG and starting graduate students Example Applications:
DEPARTMENT OF BOTANY Telangana University Dichpally, Nizamabad -503322 (A State University Established under the Act No. 28 of 2006, A.P. Recognized by UGC under 2(f) and 12 (B) of UGC Act 1956)
