A SUPERVISED FACTORIAL ACOUSTIC MODEL FOR SIMULTANEOUS MULTIPARTICIPANT VOCAL ACTIVITY DETECTION IN CLOSE-TALK MICROPHONE RECORDINGS OF MEETINGS

Using automatic speech recognition (ASR) word error rates ( WERs) as a metric, the systems in (1) and (3) appear to have yiel d similar performance, in spite of significant additional arc hitectural differences. Systems of type (2) have not been fie lded for segmentation for ASR, and therefore cannot be directly comp ared. Although approaches of type (3) offer a significant advantag e, namely the opportunity to directly constrain the number o f simultaneously vocalizing participants, they come with the c aveat of a variable acoustic vector size, since conversatio ns/meetings can have variable numbers of participants. To overcome this difficulty, unsupervised acoustic models have been deploye d [4], which do not require acoustic model training data (or traini ng time). Our previous work has shown that this severly limit s the number of features, as well as the minimum frame size. The aim of the current work is to develop a supervised acoustic model , capable of producing accurate density estimates for large f eature vectors extracted from short frames, for scenario (3 ).