Enhancements for Audio-only Diarization Systems
This work addresses the challenging clustering problem in speaker diarization for applications like meeting transcription, but it is incremental as it builds on existing methods and assumes known speaker counts and perfect segmentation.
The paper tackles the speaker clustering component of audio-only diarization systems by proposing enhancements like temporal smoothing and improvements to Deep Embedded Clustering, achieving relative improvements of 20% on AMI and 19% on DIHARD tasks when the number of speakers is known.
In this paper two different approaches to enhance the performance of the most challenging component of a Speaker Diarization system are presented, i.e. the speaker clustering part. A processing step is proposed enhancing the input features with a temporal smoothing process combined with nonlinear filtering. We, also, propose improvements on the Deep Embedded Clustering (DEC) algorithm -- a nonlinear feature transformation. The performance of these enhancements is compared with different clustering algorithms, such as the UISRNN, k-Means, Spectral clustering and x-Means. The evaluation is held on three different tasks, i.e. the AMI, DIHARD and an internal meeting transcription task. The proposed approaches assume a known number of speakers and time segmentations for the audio files. Since, we focus only on the clustering component of diarization for this work, the segmentation provided is assumed perfect. Finally, we present how supervision, in the form of given speaker profiles, can further improve the overall diarization performance. The proposed enhancements yield substantial relative improvements in all 3 tasks, with 20\% in AMI and 19\% better than the best diarization system for DIHARD task, when the number of speakers is known.