arXiv subjects

Juan Ignacio Alvarez-Trejos

Publications and source records attributed to Juan Ignacio Alvarez-Trejos.

3 recordsLinked to original sources

arXiv2026

DASHA: Diarization-Aware Speech-LLM ASR for Healthcare in Code-Switched Indic Doctor-Patient Conversations

Extracting patients' medical conditions from recorded primary-care conversations requires knowing who said what, which is difficult when speech is far-field, spontaneous, overlapping and code-switched. We study this problem on DISPLACE-M, a corpus of Hindi--English (Hinglish) doctor--patient conversations, and propose DASHA, a modular cascade of diarization, speaker-attributed ASR (SA-ASR) and LLM-based extraction. An end-to-end neural diarization with vector clustering (EEND-VC) system built on a multilingual w2v-BERT~2.0 encoder reduces DER from 9.31% to 7.76%. A Qwen3-ASR model adapted to Hindi and the target domain, with Devanagari normalization, reduces tcpWER from 26.78% to 19.27%, and to 18.59% with optional LLM correction. A module-swap analysis shows that better diarization improves extraction only when paired with the adapted ASR. On code-switched words, the multilingual encoder roughly halves speaker confusion. DASHA sets a new state of the art for two-speaker, code-switched Hindi SA-ASR and ranked first among 25 participants in the DISPLACE-M challenge.

eess.AS↗

arXiv2025

Probabilistic Fusion and Calibration of Neural Speaker Diarization Models

End-to-End Neural Diarization (EEND) systems produce frame-level probabilistic speaker activity estimates, yet since evaluation focuses primarily on Diarization Error Rate (DER), the reliability and calibration of these confidence scores have been largely neglected. When fusing multiple diarization systems, DOVER-Lap remains the only established approach, operating at the segment level with hard decisions. We propose working with continuous probability outputs, which enables more sophisticated fusion and calibration techniques that can leverage model uncertainty and complementary strengths across different architectures. This paper presents the first comprehensive framework for calibrating and fusing EEND models at the probability level. We investigate two output formulations (multilabel and powerset representations) and their impact on calibration and fusion effectiveness. Through extensive experiments on the CallHome two-speaker benchmark, we demonstrate that proper calibration provides substantial improvements even for individual models (up to 19% relative DER reduction), in some cases mitigating the absence of domain adaptation. We reveal that joint calibration in powerset space consistently outperforms independent per-speaker calibration, that fusion substantially improves over individual models, and that the Fuse-then-Calibrate ordering generally outperforms both calibrating before fusion and uncalibrated fusion while requiring calibration of only a single combined model. Our best configuration outperforms DOVER-Lap in terms of DER while providing reliable confidence estimates essential for downstream applications. This work proposes best practices for probability-level fusion of EEND systems and demonstrates the advantages of leveraging soft outputs over hard decisions.

cs.SD↗

arXiv2024

Leveraging Speaker Embeddings in End-to-End Neural Diarization for Two-Speaker Scenarios

End-to-end neural speaker diarization systems are able to address the speaker diarization task while effectively handling speech overlap. This work explores the incorporation of speaker information embeddings into the end-to-end systems to enhance the speaker discriminative capabilities, while maintaining their overlap handling strengths. To achieve this, we propose several methods for incorporating these embeddings along the acoustic features. Furthermore, we delve into an analysis of the correct handling of silence frames, the window length for extracting speaker embeddings and the transformer encoder size. The effectiveness of our proposed approach is thoroughly evaluated on the CallHome dataset for the two-speaker diarization task, with results that demonstrate a significant reduction in diarization error rates achieving a relative improvement of a 10.78% compared to the baseline end-to-end model.

cs.SD↗