Multispeaker Localization and Tracking in Intelligent Environments

Automatic speaker localization is an important task in several applications such as acoustic scene analysis, hands-free videoconferencing orspeechenhancement.Trackingspeakersinmultipartyconversationsconstitutes a fundamental task for automatic meeting analysis. In this work, we present the acoustic Person Tracking system developed at the UPC for the CLEAR’07 evaluation campaign. The designed system is able to track the estimated position of multiple speakers in a smart-room environment. Preliminary speaker locations are provided by the SRP-PHAT algorithm, which is known to perform robustly in most scenarios. Data association techniques based on trajectory prediction and spatizal clustering are used to match the raw positional estimates with potential speakers. These positional measurements are then finally spatially smoothed by means of Kalman filtering. Besides the technology description, experimental results obtained on the CLEAR’07 CHIL database are also reported.

• Type: Conference
• Year: 2007
• Where: CLEAR
• Authors: Carlos Segura, Alberto Abad, Javier Hernando, Climent Nadeu
• Keywords: Acoustic Scene Analysis, Automatic Speaker Localization, Biometrics, CLEAR 2007, Raw Positional Estimates