Automatic Classification and Transcription of Telephone Speech in Radio Broadcast Data

Abstract

Abstract. Automatic transcription of telephone speech involves additional challenges compared to wideband data processing, mainly due to channel limitations and to particular characteristics of conversational telephone speech. While in TV speech recognition applications, such as automatic transcription of broadcast news, the presence of telephone data is nearly insignificant (less than 1 %), in most radio broadcast stations the presence of telephone speech grows significantly. Thus, transcription of telephone speech data deserves special attention in radio broadcast applications. In this work, we describe our initial efforts to tackle this particular problem. First, a telephone channel classifier is proposed to automatically detect telephone segments. Then, some strategies for increasing robustness of the automatic transcription system are investigated.

Automatic word naming recognition for an on-line aphasia treatment system

“…The version of AUDIMUS integrated in VITHEA uses an acoustic model trained with 57 h of downsampled Broadcast News data and 58 h of mixed fixed-telephone and mobile-telephone data in European Portuguese (Abad and Neto, 2008). The number of context input frames is 13 for the PLP and RASTA networks and 15 for the MSG network.…”

Speech and language technologies for the automatic monitoring and training of cognitive functions

“…The baseline system combines three MLP-based acoustic models trained with Perceptual Linear Prediction features (PLP, 13 static + first derivative), log-RelAtive SpecTrAl features (RASTA, 13 static + first derivative) and Modulation SpectroGram features (MSG, 28 static). These model networks were trained with 57 hours of downsampled Broadcast News data and 58 hours of mixed fixed-telephone and mobile-telephone data in European Portuguese [19]. The number of context input frames is 13 for the PLP and RASTA networks and 15 for the MSG network.…”