This work presents first results in segmenting and classifying an Italian audio broadcast news corpus, under development at ITC-irst. The approach is based on two stages: during the first one, a HMM decoder segments and classifies the input audio stream in terms of acoustic source (music, speech, speech in presence of music). The second stage, based on the BIC method, detects speaker changes inside individual speech segments obtained through the first stage. On the test set, frame classification accuracy is 90.6%, while recall and precision of spectral changes detection with a tolerance of +-300cs are 82.9% and 65.6%, respectively
Segmentation and Classification of Italian Audio Broadcast News
Cettolo, Mauro
1999-01-01
Abstract
This work presents first results in segmenting and classifying an Italian audio broadcast news corpus, under development at ITC-irst. The approach is based on two stages: during the first one, a HMM decoder segments and classifies the input audio stream in terms of acoustic source (music, speech, speech in presence of music). The second stage, based on the BIC method, detects speaker changes inside individual speech segments obtained through the first stage. On the test set, frame classification accuracy is 90.6%, while recall and precision of spectral changes detection with a tolerance of +-300cs are 82.9% and 65.6%, respectivelyFile in questo prodotto:
Non ci sono file associati a questo prodotto.
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
