Abstract: This paper presents the speaker diarisation system presented by Aholab Signal Processing Laboratory to the Albayzin Speaker Diarization Challenge 2010. The system was built to run online, without any recording of the audio data to produce its output. As a result, the whole process must be done in a single iteration, which prevents the use of many optimisation processes that are usually implemented in diarisation systems. In order to minimise the reduction of the accuracy in the output and to maximise computational efficiency, the applied algorithms were carefully selected and some new modifications were implemented.
Index Terms: speaker diarisation, BIC, online audio processing.