Deep Learning based Identification of Primary Speaker in Voice-Controlled Devices

摘要

At present, so many low budget recording devices available, recording lectures, meetings, conferences and events have become a very easy option for everyone but also at the same time these devices lead to unclear speeches. An advanced methodology of identifying the primary speaker is introduced to record noisy audio file given in any regional language. The speaker turns and speaker lengths are examples of features which provide greater insight in the detection of primary speakers. This also provides the transcript of the primary speaker audio chunk to the end-user. Speaker diarization is the process of identifying various chunks in given audio belonging to different homogenous speakers where the count of speakers is unknown. This process is a mixture of segmentation and clustering. Speech segmentation detects the speaker change points followed by grouping them based on the speaker. Thus, Speaker Diarization is the most important step in the Identification of the primary speaker.