@inproceedings{mitra-etal-2026-llm,
title = "From {LLM} Prompts to Acoustic Baselines: A Scalable Pipeline for Under-Resourced Disfluent Code-Mixed Speech",
author = "Mitra, Anuran and
Chakravarty, Anirvan and
Mondal, Tapabrata and
Bandyopadhyay, Sivaji",
editor = "Ojha, Atul Kr. and
Sakti, Sakriani and
Soria, Claudia and
Melero, Maite and
McCrae, John P. and
Lignos, Constantine and
Liu, Chao-Hong and
Claramunt, German Rigau and
Rehm, Georg",
booktitle = "Proceedings of the {SIGUL} 2026 Joint Workshop with {ELE}, {EURALI}, and {DCLRL}: Towards Inclusivity and Equality: Language Resources and Technologies for Under-Resourced and Endangered Languages",
month = may,
year = "2026",
address = "Palma, Mallorca, Spain",
publisher = "ELRA Language Resources Association (ELRA)",
url = "https://aclanthology.org/2026.sigul-1.22/",
doi = "10.63317/5k9xcfdyrtev",
pages = "222--232",
abstract = "Spontaneous speech in multilingual communities often involves rapid code-mixing (CM) and natural disfluencies, yet such patterns are rarely reflected in available training data for under-resourced languages. This gap limits the development of robust automatic speech recognition (ASR) systems. To address this, we introduce BEHE-CMDisfl, a fully synthetic Bengali{--}English and Hindi{--}English disfluent code-mixed speech corpus generated through a controlled Large Language Model (LLM) and Text-to-Speech (TTS) pipeline. The dataset explicitly incorporates conversational phenomena such as filled pauses, repetitions, and restarts. We evaluate its usefulness under two ASR settings. In a micro-resource scenario ({\ensuremath{\sim}}1.3 hours), a GMM-HMM Kaldi baseline achieved a 37.74{\%} Word Error Rate (WER) after phonetic normalization to reduce transliteration inconsistencies, and successfully retained disfluency markers in decoding. We also examined adaptation of a modern foundation model. In zero-shot testing, openai/whisper-small failed on the code-mixed speech due to severe hallucinations and looping behavior. After applying parameter-efficient fine-tuning (LoRA) for 1,000 steps, the model stabilized, reduced insertion errors, captured rapid language switching more reliably, and achieved a WER of 21.37{\%}. These findings show that synthetic data combined with efficient fine-tuning offers a practical path for ASR development in complex low-resource disfluent CM settings."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="mitra-etal-2026-llm">
<titleInfo>
<title>From LLM Prompts to Acoustic Baselines: A Scalable Pipeline for Under-Resourced Disfluent Code-Mixed Speech</title>
</titleInfo>
<name type="personal">
<namePart type="given">Anuran</namePart>
<namePart type="family">Mitra</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Anirvan</namePart>
<namePart type="family">Chakravarty</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tapabrata</namePart>
<namePart type="family">Mondal</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sivaji</namePart>
<namePart type="family">Bandyopadhyay</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-05</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the SIGUL 2026 Joint Workshop with ELE, EURALI, and DCLRL: Towards Inclusivity and Equality: Language Resources and Technologies for Under-Resourced and Endangered Languages</title>
</titleInfo>
<name type="personal">
<namePart type="given">Atul</namePart>
<namePart type="given">Kr.</namePart>
<namePart type="family">Ojha</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sakriani</namePart>
<namePart type="family">Sakti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Claudia</namePart>
<namePart type="family">Soria</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Maite</namePart>
<namePart type="family">Melero</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">John</namePart>
<namePart type="given">P</namePart>
<namePart type="family">McCrae</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Constantine</namePart>
<namePart type="family">Lignos</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chao-Hong</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">German</namePart>
<namePart type="given">Rigau</namePart>
<namePart type="family">Claramunt</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Georg</namePart>
<namePart type="family">Rehm</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>ELRA Language Resources Association (ELRA)</publisher>
<place>
<placeTerm type="text">Palma, Mallorca, Spain</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Spontaneous speech in multilingual communities often involves rapid code-mixing (CM) and natural disfluencies, yet such patterns are rarely reflected in available training data for under-resourced languages. This gap limits the development of robust automatic speech recognition (ASR) systems. To address this, we introduce BEHE-CMDisfl, a fully synthetic Bengali–English and Hindi–English disfluent code-mixed speech corpus generated through a controlled Large Language Model (LLM) and Text-to-Speech (TTS) pipeline. The dataset explicitly incorporates conversational phenomena such as filled pauses, repetitions, and restarts. We evaluate its usefulness under two ASR settings. In a micro-resource scenario (\ensuremath\sim1.3 hours), a GMM-HMM Kaldi baseline achieved a 37.74% Word Error Rate (WER) after phonetic normalization to reduce transliteration inconsistencies, and successfully retained disfluency markers in decoding. We also examined adaptation of a modern foundation model. In zero-shot testing, openai/whisper-small failed on the code-mixed speech due to severe hallucinations and looping behavior. After applying parameter-efficient fine-tuning (LoRA) for 1,000 steps, the model stabilized, reduced insertion errors, captured rapid language switching more reliably, and achieved a WER of 21.37%. These findings show that synthetic data combined with efficient fine-tuning offers a practical path for ASR development in complex low-resource disfluent CM settings.</abstract>
<identifier type="citekey">mitra-etal-2026-llm</identifier>
<identifier type="doi">10.63317/5k9xcfdyrtev</identifier>
<location>
<url>https://aclanthology.org/2026.sigul-1.22/</url>
</location>
<part>
<date>2026-05</date>
<extent unit="page">
<start>222</start>
<end>232</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T From LLM Prompts to Acoustic Baselines: A Scalable Pipeline for Under-Resourced Disfluent Code-Mixed Speech
%A Mitra, Anuran
%A Chakravarty, Anirvan
%A Mondal, Tapabrata
%A Bandyopadhyay, Sivaji
%Y Ojha, Atul Kr.
%Y Sakti, Sakriani
%Y Soria, Claudia
%Y Melero, Maite
%Y McCrae, John P.
%Y Lignos, Constantine
%Y Liu, Chao-Hong
%Y Claramunt, German Rigau
%Y Rehm, Georg
%S Proceedings of the SIGUL 2026 Joint Workshop with ELE, EURALI, and DCLRL: Towards Inclusivity and Equality: Language Resources and Technologies for Under-Resourced and Endangered Languages
%D 2026
%8 May
%I ELRA Language Resources Association (ELRA)
%C Palma, Mallorca, Spain
%F mitra-etal-2026-llm
%X Spontaneous speech in multilingual communities often involves rapid code-mixing (CM) and natural disfluencies, yet such patterns are rarely reflected in available training data for under-resourced languages. This gap limits the development of robust automatic speech recognition (ASR) systems. To address this, we introduce BEHE-CMDisfl, a fully synthetic Bengali–English and Hindi–English disfluent code-mixed speech corpus generated through a controlled Large Language Model (LLM) and Text-to-Speech (TTS) pipeline. The dataset explicitly incorporates conversational phenomena such as filled pauses, repetitions, and restarts. We evaluate its usefulness under two ASR settings. In a micro-resource scenario (\ensuremath\sim1.3 hours), a GMM-HMM Kaldi baseline achieved a 37.74% Word Error Rate (WER) after phonetic normalization to reduce transliteration inconsistencies, and successfully retained disfluency markers in decoding. We also examined adaptation of a modern foundation model. In zero-shot testing, openai/whisper-small failed on the code-mixed speech due to severe hallucinations and looping behavior. After applying parameter-efficient fine-tuning (LoRA) for 1,000 steps, the model stabilized, reduced insertion errors, captured rapid language switching more reliably, and achieved a WER of 21.37%. These findings show that synthetic data combined with efficient fine-tuning offers a practical path for ASR development in complex low-resource disfluent CM settings.
%R 10.63317/5k9xcfdyrtev
%U https://aclanthology.org/2026.sigul-1.22/
%U https://doi.org/10.63317/5k9xcfdyrtev
%P 222-232
Markdown (Informal)
[From LLM Prompts to Acoustic Baselines: A Scalable Pipeline for Under-Resourced Disfluent Code-Mixed Speech](https://aclanthology.org/2026.sigul-1.22/) (Mitra et al., SIGUL-EURALI-DCLRL 2026)
ACL