CONF Parthasarathi_TSD2008_2008/IDIAP Exploiting Contextual Information for Speech/Non-Speech Detection Parthasarathi, Sree Hari Krishnan Motlicek, Petr Hermansky, Hynek EXTERNAL https://publications.idiap.ch/attachments/papers/2008/Parthasarathi_TSD2008_2008.pdf PUBLIC https://publications.idiap.ch/index.php/publications/showcite/parthasarathi:rr08-22 Related documents Text, Speech and Dialogue Brno, Czech Republic Series of Lecture Notes In Artificial Intelligence (LNAI) 5246 451-459 978-3-540-87390-7 2008 Springer-Verlag Berlin, Heidelberg September 2008 In this paper, we investigate the effect of temporal context for speech/non-speech detection (SND). It is shown that even a simple feature such as full-band energy, when employed with a large-enough context, shows promise for further investigation. Experimental evaluations on the test data set, with a state-of-the-art multi-layer perceptron based SND system and a simple energy threshold based SND method, using the F-measure, show an absolute performance gain of 4.4% and 5.4% respectively. The optimal contextual length was found to be 1000 ms. Further numerical optimizations yield an improvement (3.37% absolute,',','), resulting in an absolute gain of 7.77% and 8.77% over the MLP based and energy based methods respectively. ROC based performance evaluation also reveals promising performance for the proposed method, particularly in low SNR conditions. REPORT parthasarathi:rr08-22/IDIAP Exploiting contextual information for speech/non-speech detection Parthasarathi, Sree Hari Krishnan Hermansky, Hynek EXTERNAL https://publications.idiap.ch/attachments/reports/2008/parthasarathi-idiap-rr-08-22.pdf PUBLIC Idiap-RR-22-2008 2008 IDIAP In this paper, we investigate the effect of temporal context for speech/non-speech detection (SND). It is shown that even a simple feature such as full-band energy, when employed with a large-enough context, shows promise for further investigation. Experimental evaluations on the test data set, with a state-of-the-art multi-layer perceptron based SND system and a simple energy threshold based SND method, using the F-measure, show an absolute performance gain of $4.4\%$ and $5.4\%$ respectively. The optimal contextual length was found to be 1000 ms. Further numerical optimizations yield an improvement ($3.37\%$ absolute,',','), resulting in an absolute gain of $7.77\%$ and $8.77\%$ over the MLP based and energy based methods respectively. ROC based performance evaluation also reveals promising performance for the proposed method, particularly in low SNR conditions.