Speaker Diarization Using BiLSTM and BiGRU with Self-Attention
Ananya Ayasi, Jacob Joshy, Rajeev Rajan · 2022
Speaker diarization is splitting an audio stream into segments based on speaker turns and clustering the segments according to speaker identity by using the speaker characteristics. This paper aims to devise techniques based on bidirectional Long Short Term Memory Units and Gated-Recurrent Units with self-attention for speaker segmentation in a speaker diarization system. Self-attention, also called intra-attention, is an attention mechanism relating different positions of a single sequence to compute a representation of the same sequence. Experiments were conducted on the Augmented Multi-party Interaction (AMI) Meeting Corpus, a popular speech recognition dataset. The results show the potential of BiLSTM sequential processing with attention to the speaker diarization process.