The 1998 speaker recognition evaluation is part of an ongoing series of yearly evaluations conducted by NIST. These evaluations provide an important contribution to the direction of research efforts and the calibration of technical capabilities. They are intended to be of interest to all researchers working on the general problem of text independent speaker recognition. To this end the evaluation was designed to be simple, to focus on core technology issues, to be fully supported, and to be accessible.

The 1998 evaluation will be conducted in February. A follow-up workshop for evaluation participants will be held during March, to discuss research findings. Participation in the evaluation is solicited for all sites that find the task and the evaluation of interest. For more information, and to register a desire to participate in the evaluation, please contact Dr. Alvin Martin at NIST.¹

Technical Objective

The current speaker recognition evaluation focuses on the task of speaker detection. That is, the task is to determine whether a specified target speaker is speaking during a given speech segment. ² This task is posed in the context of conversational telephone speech and for limited training data. The evaluation is designed to foster research progress, with the goals of:

1. exploring promising new ideas in speaker recognition,
2. developing advanced technology incorporating these ideas, and
3. measuring the performance of this technology.

In 1996 and 1997 handset variation was featured as a prominent technical challenge to be addressed. While handset variation remains a formidable challenge, the 1998 evaluation will direct greatest attention toward speaker recognition performance for the case in which both training and test data are from the same source.

Speaker detection performance will be evaluated by measuring the correctness of detection decisions for an ensemble of speech segments. These segments will represent a statistical sampling of conditions of evaluation interest. For each of these segments a set of target speaker identities will be assigned as a test hypotheses. Each of these hypotheses will then be required to be judged as true or false, and the correctness of these decisions will be tallied. ³

The formal evaluation measure will be a detection cost function, defined as a weighted sum of the miss and false alarm error probabilities:

The parameters of this cost function are the relative costs of detection errors, C_Miss and C_FalseAlarm, and the a priori probability of the target, P_Target. The primary evaluation will use the following parameter values:

In addition to the (binary) detection decision, a decision score will also be required for each test hypothesis.⁴ This decision score will be used to produce detection error tradeoff curves, in order see how misses may be traded off against false alarms.

In 1997, the use of telephone handsets labels (one of two types, either electret or carbon-button) was demonstrated to provide a significant improvement in speaker recognition performance. This labeling was produced automatically by analysis and classification of the same speech signal used for speaker recognition. Because of the benefit that this labeling process provides, the speech data to be provided in the 1998 evaluation will also be so labeled.⁵ During evaluation, the speaker recognition systems will be allowed access to this information, for all training and test data.

• "One-session" training. The training data will be 2 minutes of speech data taken from only one conversation. This data will be stored in two files, with 1 minute of speech in each.
• "Two-session" training. Equal amounts of training data will be taken from two different conversations collected from the same phone number. The training data for this two-session condition will comprise the first of the one-session training files, plus an additional file containing 1 minute of speech from a different session (but from the same phone number).⁶
• "Two-session-full" training. All available speech data taken from two different conversations collected from the same phone number. The training data for this two-session-full condition will comprise all the data used for the one-session and two-session training conditions, plus two additional files which will contain additional speech data available from each of the two sessions used for the other two training conditions.

• Test segment duration. Performance will be computed separately for 3 different test durations. These durations will be nominally 3 seconds, 10 seconds and 30 seconds. Actual duration will vary from nominal so that whole turns may be included whenever possible. Actual durations will be constrained to be within the ranges of 2-4 seconds, 7-13 seconds, and 25-35 seconds, respectively. A single turn will be used for the test segments whenever possible.
• Same/different phone number.⁷ Performance will be computed separately for test segments from the training phone number versus those segments from different phone numbers. For this test, the handset type label (electret or carbon-button) will be the same as that used in training, and it will be provided as side information to the system under test.
• Same/different handset type. Performance will be computed separately for test segments with the same handset type label as training, versus segments with a different handset label. For this test, all test segments will be from phone numbers different from the training number.

The development data for this evaluation will be the EvalSet for the 1997 evaluation. The 1997 EvalSet exists on six CD-ROM's labeled sid97e*. Sites intending to perform the evaluation and to submit results to NIST may acquire these data and associated documentation from NIST free of charge by contacting Dr. Martin.⁸

Evaluation Data

The evaluation data will be drawn from the SwitchBoard-2 phase 2 corpus.⁹ Both training and test segments will be constructed by concatenating consecutive turns for the desired speaker, similar to what was done last year (including, processing each conversation through echo cancelling software before each test segment is created). Each segment will be stored as a continuous speech signal in a separate SPHERE file.¹⁰ The speech data will be stored in 8-bit m -law format. The handset type label will be supplied in the SPHERE header. The SPHERE header will also include auxiliary information to document to source file, start time and duration of all excerpts which were used to construct the segment.

NIST will verify that the speech segments are properly labeled and fairly represent the speaker. There will be about 250 female and 250 male speakers. These speakers will serve both as target speakers and as non-target (impostor) speakers.¹¹

The evaluation corpus will be supplied on 6 CD-ROM's, by necessity. For convenience, data will be grouped according to sex and stored separately - three discs for female data and three discs for male data. Knowledge of the sex of the target speaker is admissible side information and may be used if desired.

The evaluation data will include both training data and test data. There will be an average of about 10 test segments for each target speaker and for each test duration. (Each of these test segments will be from a different conversation.) This will make a total of about 2500 test segments for each sex and for each of the three test durations. (Note, however, that some of the test segments will be from speakers other than the target speakers.) For each test segment, there will be about ten trials. Thus, for all of the tests in this evaluation, there will be a total of about 45,000 target speaker trials and about 405,000 non-target speaker trials. Giving a grand-total of about 450,000 tests [ 2500 files/duration/sex * 2 sexes * 3 durations * 3 training conditions * 10 tests per file ]

A total of nine tests constitute the evaluation. These tests are namely a test for each of the three test durations for each of the three training conditions. Every evaluation participant is required to submit all of the results for each test performed.¹² In the event that a participating site does not submit a complete set of results, NIST will not report any results for that site.

The following evaluation rules and restrictions must be observed by all participants:

• Each decision is to be based only upon the specified test segment and target speaker. Use of information about other test segments and/or other target speakers is not allowed.¹³ For example,
• Normalization over multiple test segments is not allowed.
• Normalization over multiple target speakers is not allowed.
• Use of evaluation data for impostor modeling is not allowed.
• The use of transcripts for target speaker training is not allowed.
• Knowledge of the training conditions (implied by data set directory structure as indicated below) is allowed.
• Knowledge of the sex of the target speaker (implied by data set directory structure as indicated below) is allowed.
• Knowledge of the handset type (found in the SPHERE header under the field name "handset_type" -- see SPHERE Header Information, below) is allowed.
• Listening to the evaluation data, or any other experimental interaction with the data, is not allowed before all test results have been submitted. This applies to target speaker training data as well as test segments.
• The corpus from which the evaluation data are taken, namely the SwitchBoard-2 phase 2 corpus, may not be used for any system training or R&D activities related to this evaluation.¹⁴
• Knowledge of the "start and ending" times that were used to construct the test segments (found in the SPHERE header -- see SPHERE Header Information, below) is allowed.
• Knowledge of the test condition: same phone number versus different phone number test (given in the index files, as described below), is allowed.
• Knowledge of any information available in the SPHERE header is allowed.

All of the six disks in the EvalSet will have the same organization. Each disk's directory structure will organize the data according to information admissible to the speaker recognition system. Some directories on some disks may be empty. This directory structure will be as follows:

• There will be a single top-level directory on each disk, used as a unique label for the disk. These directories will be named sid98e1f, sid98e2f, and sid98e3f for the female data, and sid98e1m, sid98e2m, and sid98e3m for the male data.
• Under each top-level directory there will be two data subdirectories, namely train for storing the training data and test for storing the test data, and one documenation directory doc.
  • Under the train directory there will be five subdirectories, namely s1a (for the first one-session training segment), s1b (for the second one-session training segment), s2a (for the second two-session training segment) (on the e1" discs only), s1r (for the additional two-session-full training data from the first session) and s2r (for the additional two-session-full training data from the second session) (on the "e2" discs only)
  • In each of these directories there will be one SPHERE-format speech data file for each of the speakers, containing the appropriate amount speech in m-law format. The name of this file will be the ID of the target speaker, followed by ".sph".
  • Under the test directory there will be three subdirectories, namely " 30" (for the 30 second test segments), "10" (for the 10 second test segments), and "3" (for the 3 second test segments).
  • In each of the 30, 10 and 3 segment duration directories will be the test segments of that duration, one SPHERE-format m -law speech data file for each test segment. The name of these files will be pseudo-random alphanumeric strings, followed by ".sph".

• 1s.ndx (for targets using the one-session training condition)
• 2s.ndx (for targets using the two-session training condition)
• 2f.ndx (for targets using the two-session-full training condition)

Sites participating in the evaluation must report test results for all of the tests. These results must be provided to NIST in results files using a standard ASCII record format, with one record for each decision. Each record must document its decision with target identification, test segment identification, and decision information. Each record must thus contain seven fields, separated by white space and in the following order:

1. The sex of the target speaker - M or F.
2. The training condition - 1S or 2S or 2F. (one-session or two-session or two-session-full)
3. The target speaker ID. (a 4-digit number)
4. The test segment duration - 30, 10 or 3.
5. The test segment file name. (excluding the directory and file type)
6. The decision - T or F. (Is the target speaker the same as the test segment speaker?)
7. The score. (where the more positive the score, the more likely the target speaker)

A brief description of the system (the algorithms) used to produce the results must be submitted along with the results, for each system evaluated. (It is permissible for a single site to submit multiple systems for evaluation. In this case, however, the submitting site must identify one system as the "primary" system prior to performing the evaluation.)

Sites must report the CPU execution time that was required to process the test data, as if the test were run on a single CPU. Sites must also describe the CPU and the amount of memory used.

• The deadline for signing up to participate in the evaluation is 16 January 1998.
• The evaluation data set CD-ROM's will be distributed by NIST on 9 February 1998.
• The deadline for submission of evaluation results to NIST is 23 February 1998.
• Room reservations for the follow-up workshop (see below) must be received by March 9, 1998.
• The follow-up workshop will be held at The Inn and Conference Center of the University of Maryland University College in College Park, Maryland on 31 March and 1 April 1998¹⁶. Participants in the evaluation will be expected to attend this workshop and to present and discuss their research at it.

SPHERE Header Information