This demo shows the FBK Speaker Localization and Tracking real-time system based on a distributed microphone network. The system is based on the measurement in the space of a score that represents the plausibility that a sound source is in a given 3D point in space (Global Coherence Field, derived from a consolidated technique, ITC-irst 1992).
You are here
Overview on DEMOS page
This demo shows the FBK Acoustic Event Detection and Classification System. The real-time prototype is able to recognize a set of 12 different acoustic events. Some examples of events are: door knocking or slamming, phone ringing, laughs, coughs, and obviously speech.
This demo shows the FBK real-time system for detection of speech activity using close-talk microphones, given the fact that the speakers overlap in time one each other.
This demo shows FBK real-time system for Far Field Connected Digit Recognition. The system is based on the input from one of the far-field microphones installed on the wall opposite to the speakers, that is with a speaker-microphone distance ranging between 3 and 5 meters.
This demo shows the FBK Fundamental Frequency Extraction system based on 14 far-field microphones. The system is able to extract the pitch of male and female voices as well as of singing voices and of musical instruments.
Real-time Blind Source Extraction demonstration presented at Interspeech 2011. Three multiple interfering sources and a target speaker, uttering some commands, are recorded by two microphones. The BSE extracts only the target source signal while filtering out the remaining noise. The extracted signal is fed to an ASR system and the recognition output is visualized.
This video shows an interesting feature of the joint audio-video tracking tool developed at FBK by SHINE and TEV units. When light goes off the system can still track the target thanks to the joint use of audio and video information.
This video shows a demo of the joint Audio-Video multitarget tracking system developed jointly by SHINE and TEV units at FBK. Bubbles appear when some speech activity is detected for a target and audio likelihood is included in the tracking process.
Demo developed at FBK (Shine unit)- Tracking multiple acoustic sources using a distributed microphone network.
A short clip describing the MultiPath-GCF (MP-GCF): an algorithm for the localization of acoustic sources, based on multipath audio propagation.
This video shows a demo of the 2D Real-time Tracking of Multiple Acoustic Sources -Semi Blind Source Separation, GSCT, TDOA.