The World Wide Web Consortium (W3C) held a workshop on speaker identification and verification (SIV) in Menlo Park, Calif., in March. The purpose of this workshop was to identify and prioritize directions for SIV standards work as a means of making SIV more useful in current and emerging markets. Prior to the meeting, each participant had submitted a position paper supporting one or more of the following workshop goals:
- Identify application requirements for SIV in VoiceXML 3.0;
- Identify SIV standards relevant to VoiceXML 3.0; and
- Integrate existing and in-process standards with VoiceXML 3.0.
Fifteen participants from the United States, Germany, Australia, Japan, and the United Kingdom assembled at the Menlo Park facility of SRI International, which sponsored the event.
The participants represented primary SIV stakeholders: customers (General Motors, CentreLink), vendors (Nuance Communications, Recognition Technologies, SRI International), integrators and platform developers (Cisco, Daon, Deutsche Telekom, iBiometrics, Voxeo), and consultants (EIG Technologies; J. Markowitz Consultants). Also present were the editor of the Internet Engineering Technology Forum’s Media Resources Control Protocol (MRCP v2), the chair of the OASIS Biometric Identity Assurance Services (BIAS) Integration Technical Committee, the editor of the American National Standards Institute’s (ANSI’s) draft standard on voice data interchange, the chair of the W3C’s Voice Browser Working Group (VBWG), several members of the VoiceXML Forum’s Speaker Biometrics Committee (SBC), and several members of the W3C Voice Browser Working Group. They agreed that the SIV market is broad, has strong public and private-sector components, and that development of the SIV module should be predicated on the needs and issues of real-world deployments.
In addition to considering the core functionality needed for an SIV module of VoiceXML, discussions covered security, markets, and integration with other standards. The participants were particularly concerned about the security aspects of SIV. One of the products of the workshop was an architecture for shielding sensitive data that was inspired by the high-security solution deployed by Australia’s CentreLink (social services) agency. A related outcome was a directive to the SBC to do further development of security guidelines.
Considerations of functionality extended beyond the core enroll, verify, and identify operations to issues regarding where biometric matching and decision-making are performed, multimodal and multimedia integration, and whether to include administrative functions in SIV module. The participants in the Workshop recommended that when the VBWG works on the SIV module, the VBWG should reference existing work rather than re-creating it. This involves coordination with other organizations, notably ANSI/INCITS M1 (biometrics technical committee), OASIS/BIAS, and the SBC. Although speaker identification and classification were noted as areas of interest, the technical discussions focused primarily on verification.
The architecture, the SBC mandate, a list of reasons why SIV should be added to VoiceXML, and a list of other points of consensus will be presented to the W3C’s VBWG, which is responsible for developing the SIV specification for use within VoiceXML 3.0. These materials and a summary of the workshop sessions will also be published on the SIV workshop page of the W3C’s Web site.
Judith Markowitz, Ph.D., is president of J. Markowitz Consultants and a leading independent analyst in the speech and voice biometrics field. She can be reached at email@example.com.