@inproceedings{chiriatti-etal-2026-gold,
title = "Gold-Preserving Synthetic Augmentation for Verbatim {S}wedish Automatic Speech Recognition",
author = "Chiriatti, Giuseppe and
Kokkinakis, Dimitrios and
Bertini, Flavio and
Mu{\~n}oz S{\'a}nchez, Ricardo and
Lange, Herbert",
editor = "T.Y.S.S., Santosh and
Ghosh, Sreyan and
Duraiswami, Ramani and
Watanabe, Shinji and
Arora, Siddhant and
Kumar, Sonal and
Adi, Yossi and
Manocha, Dinesh and
Garcia Perera, Leibny Paola",
booktitle = "The Proceedings of Speech and Audio Language Models Workshop (2nd Edition)",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.salma-1.4/",
pages = "38--44",
ISBN = "979-8-89176-496-5",
abstract = "Verbatim transcripts, which keep the hesitations and interactional particles that ordinary automatic speech recognition (ASR) removes, are a prerequisite for speech-based cognitive assessment, and they are hardest to obtain for languages with little annotated speech. We study this problem for Swedish with a 1.02-hour corpus of manually transcribed spontaneous speech, too small for full fine-tuning of KB-Whisper-large. We adapt the model with a low-rank adapter and compare three ways of extending the training data: pseudo-labeling of untranscribed in-domain audio, unguided synthesis with a language model and a text-to-speech system, and a gold-preserving pipeline that reproduces the disfluency distribution of the gold corpus by construction and can be audited before training. Fine-tuning reduces disfluency-segment word error rate (WER) by 32{\%} relative, with blockwise-bootstrap confidence intervals disjoint from the zero-shot baseline, while the three augmentation strategies are not separable by WER across five seeds. The gold-preserving pipeline needs no untranscribed audio, and its disfluency distribution can be verified in advance."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="chiriatti-etal-2026-gold">
<titleInfo>
<title>Gold-Preserving Synthetic Augmentation for Verbatim Swedish Automatic Speech Recognition</title>
</titleInfo>
<name type="personal">
<namePart type="given">Giuseppe</namePart>
<namePart type="family">Chiriatti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dimitrios</namePart>
<namePart type="family">Kokkinakis</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Flavio</namePart>
<namePart type="family">Bertini</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ricardo</namePart>
<namePart type="family">Muñoz Sánchez</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Herbert</namePart>
<namePart type="family">Lange</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>The Proceedings of Speech and Audio Language Models Workshop (2nd Edition)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Santosh</namePart>
<namePart type="family">T.Y.S.S.</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sreyan</namePart>
<namePart type="family">Ghosh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ramani</namePart>
<namePart type="family">Duraiswami</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shinji</namePart>
<namePart type="family">Watanabe</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Siddhant</namePart>
<namePart type="family">Arora</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sonal</namePart>
<namePart type="family">Kumar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yossi</namePart>
<namePart type="family">Adi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dinesh</namePart>
<namePart type="family">Manocha</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Leibny</namePart>
<namePart type="given">Paola</namePart>
<namePart type="family">Garcia Perera</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">979-8-89176-496-5</identifier>
</relatedItem>
<abstract>Verbatim transcripts, which keep the hesitations and interactional particles that ordinary automatic speech recognition (ASR) removes, are a prerequisite for speech-based cognitive assessment, and they are hardest to obtain for languages with little annotated speech. We study this problem for Swedish with a 1.02-hour corpus of manually transcribed spontaneous speech, too small for full fine-tuning of KB-Whisper-large. We adapt the model with a low-rank adapter and compare three ways of extending the training data: pseudo-labeling of untranscribed in-domain audio, unguided synthesis with a language model and a text-to-speech system, and a gold-preserving pipeline that reproduces the disfluency distribution of the gold corpus by construction and can be audited before training. Fine-tuning reduces disfluency-segment word error rate (WER) by 32% relative, with blockwise-bootstrap confidence intervals disjoint from the zero-shot baseline, while the three augmentation strategies are not separable by WER across five seeds. The gold-preserving pipeline needs no untranscribed audio, and its disfluency distribution can be verified in advance.</abstract>
<identifier type="citekey">chiriatti-etal-2026-gold</identifier>
<location>
<url>https://aclanthology.org/2026.salma-1.4/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>38</start>
<end>44</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Gold-Preserving Synthetic Augmentation for Verbatim Swedish Automatic Speech Recognition
%A Chiriatti, Giuseppe
%A Kokkinakis, Dimitrios
%A Bertini, Flavio
%A Muñoz Sánchez, Ricardo
%A Lange, Herbert
%Y T.Y.S.S., Santosh
%Y Ghosh, Sreyan
%Y Duraiswami, Ramani
%Y Watanabe, Shinji
%Y Arora, Siddhant
%Y Kumar, Sonal
%Y Adi, Yossi
%Y Manocha, Dinesh
%Y Garcia Perera, Leibny Paola
%S The Proceedings of Speech and Audio Language Models Workshop (2nd Edition)
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-496-5
%F chiriatti-etal-2026-gold
%X Verbatim transcripts, which keep the hesitations and interactional particles that ordinary automatic speech recognition (ASR) removes, are a prerequisite for speech-based cognitive assessment, and they are hardest to obtain for languages with little annotated speech. We study this problem for Swedish with a 1.02-hour corpus of manually transcribed spontaneous speech, too small for full fine-tuning of KB-Whisper-large. We adapt the model with a low-rank adapter and compare three ways of extending the training data: pseudo-labeling of untranscribed in-domain audio, unguided synthesis with a language model and a text-to-speech system, and a gold-preserving pipeline that reproduces the disfluency distribution of the gold corpus by construction and can be audited before training. Fine-tuning reduces disfluency-segment word error rate (WER) by 32% relative, with blockwise-bootstrap confidence intervals disjoint from the zero-shot baseline, while the three augmentation strategies are not separable by WER across five seeds. The gold-preserving pipeline needs no untranscribed audio, and its disfluency distribution can be verified in advance.
%U https://aclanthology.org/2026.salma-1.4/
%P 38-44
Markdown (Informal)
[Gold-Preserving Synthetic Augmentation for Verbatim Swedish Automatic Speech Recognition](https://aclanthology.org/2026.salma-1.4/) (Chiriatti et al., SALMA 2026)
ACL