@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://preview.aclanthology.org/cawl-year/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."
}