An Automatic Speech Recognition solution with speaker identification support

Automatic Speech Recognition may suffer in terms of intelligibility if the audio recording contains heterogeneous regions of multiple speakers, music or noise. Diarization is the process of segmenting an audio file into homogeneous regions and when used in conjunction with an Automatic Speech Recognition system, it filters out the non-speech audio regions and significantly improves the intelligibility of the recognition output. In this paper, we present an integrated diarization and transcription solution for the Romanian Language. The solution implements the diarization component as a processing stage in the speech recognizer front end. The integrated system is evaluated in terms of computation efficiency and transcription intelligibility.