MIXER

The Mixer project promotes the development of robust speaker recognition technology by providing speech data, carefully collected and audited, in which each of a large pool of speakers is recorded simultaneously across numerous microphones, and in different communicative situations and/or in multiple languages. The success of the original collection has lead to follow-on collections. Mixer data has been used for research and technology development by a number of LDC member organizations and for technology evaluation in NIST's SRE (Speaker REcognition) and LRE (Language REcognition) campaigns since 2004.

Mixer Phases

Mixer 5

The Mixer 5 targets were 300 speakers who each participated in ten calls of at least ten minutes in duration and three interview sessions. Approximately half of the subjects were interviewed at LDC and half were interviewed at ICSI. Each interview session consisted of two half-hour parts separated by a half-hour of rest. Sessions included informal conversation, more formal elicitation and up to two telephone calls. The Mixer 5 Interview Specifications describe the structure of these session in great detail. The Mixer Cross Channel Recording Setup Guide provides additional information on the configuration of the interview room and cross channel recording platform. The Mixer 5 Participant's Site shows how the study was described to potential participants.

Mixer 4

The goal of the Mixer 4 collection was to collect up to 10 telephone calls from 400 English speakers of which 2 calls form a subset of the subjects were also recorded via a cross channel recording platform that captured 14 channels of audio through a variety of microphone types. Approximately half of the subjects were recorded at LDC and half were recorded at ICSI. The Mixer Cross Channel Recording Setup Guide provides additional information on the configuration of the interview room and cross channel recording platform. The Mixer 4 Participant's Site shows how the study was described to potential participants.



Papers and Presentations

Cieri, Christopher, Linda Corson, David Graff, Kevin Walker
Resources for New Research Directions in Speaker Recognition: The Mixer 3, 4 and 5 Corpora
Interspeech 2007, Antwerp, August 2007
Available: Paper in PDF, Presentation Slides



Mixer Team

Management

  • PI: Christopher Cieri
  • Project Manager: Linda Brandschain
  • Task Coordination and Training: Abby Neely

Technical Support

  • Chris Caruso
  • David Graff
  • Kevin Walker

Data Collection

  • David Raskin
  • Nina Tchervova
  • Sophie Uy
  • John Vogel