Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TCG CREST System Description for the DISPLACE-M Challenge

Nikhil Raghav, Md Sahidullah

TL;DR

Experimental results demonstrate that the Diarizen system provides an approximate $39\% relative improvement in the diarization error rate (DER) on the post-evaluation analysis of Phase~I compared to the SpeechBrain baseline.

Abstract

This report presents the TCG CREST system description for Track 1 (Speaker Diarization) of the DISPLACE-M challenge, focusing on naturalistic medical conversations in noisy rural-healthcare scenarios. Our study evaluates the impact of various voice activity detection (VAD) methods and advanced clustering algorithms on overall speaker diarization (SD) performance. We compare and analyze two SD frameworks: a modular pipeline utilizing SpeechBrain with ECAPA-TDNN embeddings, and a state-of-the-art (SOTA) hybrid end-to-end neural diarization system, Diarizen, built on top of a pre-trained WavLM. With these frameworks, we explore diverse clustering techniques, including agglomerative hierarchical clustering (AHC), and multiple novel variants of spectral clustering, such as SC-adapt, SC-PNA, and SC-MK. Experimental results demonstrate that the Diarizen system provides an approximate $39\%$ relative improvement in the diarization error rate (DER) on the post-evaluation analysis of Phase~I compared to the SpeechBrain baseline. Our best-performing submitted system employing the Diarizen baseline with AHC employing a median filtering with a larger context window of $29$ achieved a DER of 10.37\% on the development and 9.21\% on the evaluation sets, respectively. Our team ranked sixth out of the 11 participating teams after the Phase~I evaluation.

TCG CREST System Description for the DISPLACE-M Challenge

TL;DR

Experimental results demonstrate that the Diarizen system provides an approximate $39\% relative improvement in the diarization error rate (DER) on the post-evaluation analysis of Phase~I compared to the SpeechBrain baseline.

Abstract

This report presents the TCG CREST system description for Track 1 (Speaker Diarization) of the DISPLACE-M challenge, focusing on naturalistic medical conversations in noisy rural-healthcare scenarios. Our study evaluates the impact of various voice activity detection (VAD) methods and advanced clustering algorithms on overall speaker diarization (SD) performance. We compare and analyze two SD frameworks: a modular pipeline utilizing SpeechBrain with ECAPA-TDNN embeddings, and a state-of-the-art (SOTA) hybrid end-to-end neural diarization system, Diarizen, built on top of a pre-trained WavLM. With these frameworks, we explore diverse clustering techniques, including agglomerative hierarchical clustering (AHC), and multiple novel variants of spectral clustering, such as SC-adapt, SC-PNA, and SC-MK. Experimental results demonstrate that the Diarizen system provides an approximate relative improvement in the diarization error rate (DER) on the post-evaluation analysis of Phase~I compared to the SpeechBrain baseline. Our best-performing submitted system employing the Diarizen baseline with AHC employing a median filtering with a larger context window of achieved a DER of 10.37\% on the development and 9.21\% on the evaluation sets, respectively. Our team ranked sixth out of the 11 participating teams after the Phase~I evaluation.
Paper Structure (8 sections, 3 figures, 1 table)

This paper contains 8 sections, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methodology and Experimental Setup
  3. SpeechBrain System
  4. Diarizen System
  5. Hardware requirements
  6. Result & Discussion
  7. Conclusions and Future Directions
  8. Acknowledgment

Figures (3)

  • Figure 1: Mean values of acoustic and conversational characteristics across recordings, including Speech Percentage (SP), Overlap Percentage (OVP), Absolute Difference in Pitch (ADP), Absolute Difference in F3 (ADF3), Signal-to-Noise Ratio (SNR), and Speaker Turns per Minute (STM). Error bars indicate 95% confidence intervals computed across sessions. The average values for each feature are reported in the top-right corner of the figure.
  • Figure 2: DER comparison between Diarizen and SpeechBrain systems across 78 audio files of Dev 1 set.
  • Figure 3: File-wise DER comparison sorted by $\Delta\text{DER} = \text{DER}_{\text{SpeechBrain}} - \text{DER}_{\text{Diarizen}}$. Positive values indicate improvement of Diarizen over SpeechBrain. The outlier file 3836246.wav is excluded for visualization clarity.