Guardado en:
Detalles Bibliográficos
Autores principales: Tawara, Naohiro, Cornell, Samuele, Polok, Alexander, Delcroix, Marc, Burget, Lukáš, Watanabe, Shinji
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:https://arxiv.org/abs/2603.22709
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866915886049263616
author Tawara, Naohiro
Cornell, Samuele
Polok, Alexander
Delcroix, Marc
Burget, Lukáš
Watanabe, Shinji
author_facet Tawara, Naohiro
Cornell, Samuele
Polok, Alexander
Delcroix, Marc
Burget, Lukáš
Watanabe, Shinji
contents Conversational automatic speech recognition remains challenging due to overlapping speech, far-field noise, and varying speaker counts. While recent LLM-based systems perform well on single-speaker benchmarks, their robustness in multi-speaker settings is unclear. We systematically compare LLM-based and modular pipeline approaches along four axes: overlap robustness, semantic fidelity, speaker count, and single- versus multi-channel input. To capture meaning-altering errors that conventional metrics miss, we introduce tcpSemER, which extends tcpWER by replacing Levenshtein distance with embedding-based semantic similarity. We further decompose tcpWER into overlapping and non-overlapping components for finer-grained analysis. Experiments across three datasets show that LLM-based systems are competitive in two-speaker settings but degrade as speaker count and overlap increase, whereas modular pipelines remain more robust.
format Preprint
id arxiv_https___arxiv_org_abs_2603_22709
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Who Spoke What When? Evaluating Spoken Language Models for Conversational ASR with Semantic and Overlap-Aware Metrics
Tawara, Naohiro
Cornell, Samuele
Polok, Alexander
Delcroix, Marc
Burget, Lukáš
Watanabe, Shinji
Computation and Language
Audio and Speech Processing
Conversational automatic speech recognition remains challenging due to overlapping speech, far-field noise, and varying speaker counts. While recent LLM-based systems perform well on single-speaker benchmarks, their robustness in multi-speaker settings is unclear. We systematically compare LLM-based and modular pipeline approaches along four axes: overlap robustness, semantic fidelity, speaker count, and single- versus multi-channel input. To capture meaning-altering errors that conventional metrics miss, we introduce tcpSemER, which extends tcpWER by replacing Levenshtein distance with embedding-based semantic similarity. We further decompose tcpWER into overlapping and non-overlapping components for finer-grained analysis. Experiments across three datasets show that LLM-based systems are competitive in two-speaker settings but degrade as speaker count and overlap increase, whereas modular pipelines remain more robust.
title Who Spoke What When? Evaluating Spoken Language Models for Conversational ASR with Semantic and Overlap-Aware Metrics
topic Computation and Language
Audio and Speech Processing
url https://arxiv.org/abs/2603.22709