STCON System for the CHiME-8 Challenge

TL;DR

This paper presents the STCON system for the CHiME-8 Challenge, focusing on improved diarization, speaker counting, and source separation techniques to enhance distant speech transcription accuracy.

Contribution

It introduces a carefully tuned diarization pipeline, a novel Guided Target speaker Extraction model, and data augmentation methods to improve speech recognition in multi-device recordings.

Findings

Reduced diarization error rate (DER) significantly

Achieved more reliable speech segments for recognition

Enhanced source separation with G-TSE and GSS methods

Abstract

This paper describes the STCON system for the CHiME-8 Challenge Task 1 (DASR) aimed at distant automatic speech transcription and diarization with multiple recording devices. Our main attention was paid to carefully trained and tuned diarization pipeline and speaker counting. This allowed to significantly reduce diarization error rate (DER) and obtain more reliable segments for speech separation and recognition. To improve source separation, we designed a Guided Target speaker Extraction (G-TSE) model and used it in conjunction with the traditional Guided Source Separation (GSS) method. To train various parts of our pipeline, we investigated several data augmentation and generation techniques, which helped us to improve the overall system quality.