@inproceedings{filho-etal-2026-paraspeechrag,
title = "{P}ara{S}peech{RAG}: A Benchmark for Robust Speech Retrieval under Paralinguistic and Acoustic Perturbations",
author = "Filho, Alexandre Costa Ferro and
Queiroz, Marcos Paulo C M and
Medeiros, Matheus Franco and
Sene, Gabriel Orlow De Oliveira and
Silva, Murilo Alvares and
Silva, Diogo Fernandes Costa and
Galv{\~a}o Filho, Arlindo Rodrigues",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.2581/",
pages = "58663--58674",
ISBN = "979-8-89176-459-0",
abstract = "Speech Retrieval-Augmented Generation (Speech RAG) systems are increasingly used in real-world settings where spoken queries vary across speakers, emotions, and acoustic conditions. However, current evaluations primarily focus on clean-condition retrieval accuracy, leaving unclear whether these systems remain semantically invariant under paralinguistic variation. We introduce ParaSpeechRAG, a counterfactual robustness benchmark that isolates non-semantic acoustic factors while preserving lexical content. The benchmark spans controlled perturbations across speaker identity, emotion and prosody, additive noise, speaking rate, and codec compression, enabling systematic evaluation of semantic invariance in speech retrieval. We evaluate three retriever families: ASR-based cascades, speech-text multimodal retrievers (CLASP), and general-purpose audio-language models (GLAP). Results reveal a consistent robustness{--}accuracy trade-off: multimodal retrievers achieve higher performance under clean conditions but are substantially more sensitive to paralinguistic and acoustic perturbations, particularly emotion and noise. In contrast, ASR-based cascades are more robust to these variations due to their text-centric bottleneck, though with reduced expressive capacity. These findings highlight an overlooked dimension of instability in Speech RAG systems and underscore the need for robustness-aware evaluation beyond clean-condition benchmarks."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="filho-etal-2026-paraspeechrag">
<titleInfo>
<title>ParaSpeechRAG: A Benchmark for Robust Speech Retrieval under Paralinguistic and Acoustic Perturbations</title>
</titleInfo>
<name type="personal">
<namePart type="given">Alexandre</namePart>
<namePart type="given">Costa</namePart>
<namePart type="given">Ferro</namePart>
<namePart type="family">Filho</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Marcos</namePart>
<namePart type="given">Paulo</namePart>
<namePart type="given">C</namePart>
<namePart type="given">M</namePart>
<namePart type="family">Queiroz</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Matheus</namePart>
<namePart type="given">Franco</namePart>
<namePart type="family">Medeiros</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gabriel</namePart>
<namePart type="given">Orlow</namePart>
<namePart type="given">De</namePart>
<namePart type="given">Oliveira</namePart>
<namePart type="family">Sene</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Murilo</namePart>
<namePart type="given">Alvares</namePart>
<namePart type="family">Silva</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Diogo</namePart>
<namePart type="given">Fernandes</namePart>
<namePart type="given">Costa</namePart>
<namePart type="family">Silva</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Arlindo</namePart>
<namePart type="given">Rodrigues</namePart>
<namePart type="family">Galvão Filho</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Speech Retrieval-Augmented Generation (Speech RAG) systems are increasingly used in real-world settings where spoken queries vary across speakers, emotions, and acoustic conditions. However, current evaluations primarily focus on clean-condition retrieval accuracy, leaving unclear whether these systems remain semantically invariant under paralinguistic variation. We introduce ParaSpeechRAG, a counterfactual robustness benchmark that isolates non-semantic acoustic factors while preserving lexical content. The benchmark spans controlled perturbations across speaker identity, emotion and prosody, additive noise, speaking rate, and codec compression, enabling systematic evaluation of semantic invariance in speech retrieval. We evaluate three retriever families: ASR-based cascades, speech-text multimodal retrievers (CLASP), and general-purpose audio-language models (GLAP). Results reveal a consistent robustness–accuracy trade-off: multimodal retrievers achieve higher performance under clean conditions but are substantially more sensitive to paralinguistic and acoustic perturbations, particularly emotion and noise. In contrast, ASR-based cascades are more robust to these variations due to their text-centric bottleneck, though with reduced expressive capacity. These findings highlight an overlooked dimension of instability in Speech RAG systems and underscore the need for robustness-aware evaluation beyond clean-condition benchmarks.</abstract>
<identifier type="citekey">filho-etal-2026-paraspeechrag</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2581/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>58663</start>
<end>58674</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T ParaSpeechRAG: A Benchmark for Robust Speech Retrieval under Paralinguistic and Acoustic Perturbations
%A Filho, Alexandre Costa Ferro
%A Queiroz, Marcos Paulo C. M.
%A Medeiros, Matheus Franco
%A Sene, Gabriel Orlow De Oliveira
%A Silva, Murilo Alvares
%A Silva, Diogo Fernandes Costa
%A Galvão Filho, Arlindo Rodrigues
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F filho-etal-2026-paraspeechrag
%X Speech Retrieval-Augmented Generation (Speech RAG) systems are increasingly used in real-world settings where spoken queries vary across speakers, emotions, and acoustic conditions. However, current evaluations primarily focus on clean-condition retrieval accuracy, leaving unclear whether these systems remain semantically invariant under paralinguistic variation. We introduce ParaSpeechRAG, a counterfactual robustness benchmark that isolates non-semantic acoustic factors while preserving lexical content. The benchmark spans controlled perturbations across speaker identity, emotion and prosody, additive noise, speaking rate, and codec compression, enabling systematic evaluation of semantic invariance in speech retrieval. We evaluate three retriever families: ASR-based cascades, speech-text multimodal retrievers (CLASP), and general-purpose audio-language models (GLAP). Results reveal a consistent robustness–accuracy trade-off: multimodal retrievers achieve higher performance under clean conditions but are substantially more sensitive to paralinguistic and acoustic perturbations, particularly emotion and noise. In contrast, ASR-based cascades are more robust to these variations due to their text-centric bottleneck, though with reduced expressive capacity. These findings highlight an overlooked dimension of instability in Speech RAG systems and underscore the need for robustness-aware evaluation beyond clean-condition benchmarks.
%U https://aclanthology.org/2026.emnlp-main.2581/
%P 58663-58674
Markdown (Informal)
[ParaSpeechRAG: A Benchmark for Robust Speech Retrieval under Paralinguistic and Acoustic Perturbations](https://aclanthology.org/2026.emnlp-main.2581/) (Filho et al., EMNLP 2026)
ACL