IndMedSTT-Synth (v1.0)

Misidentified medication brand names in Indian multilingual ASR transcriptions. It is part of the IndMedSTT dataset collection.

Companion dataset: IndMedSTT-Voice (coming soon)

Overview

When Indian patients describe their medications to a healthcare provider or voice-enabled systems, they typically refer to medications by their brand name, which is in English, regardless of the language they are speaking in. A multilingual Speech-to-Text (STT) model receiving this speech aggressively autocorrects the phonetically spoken brand name into the closest-sounding real English words it has been trained on, producing grammatically plausible but clinically meaningless substitutions.

For example:

• "Azithral" → "as it roll"
• "Combiflam" → "come be flame"
• "Fycare-D" → "fat cared"

This dataset provides labeled examples of such STT-corrupted transcriptions paired with their true medication brand name, medical context, and the specific misconstrued phrase. It is designed to support post-ASR medication Named Entity Recognition (NER) and ASR error correction research, particularly for Indian multilingual and code-mixed patient speech.

Dataset Details

• Size: 87 samples
• Language: English (Indian patient register)
• Source: GPT-4o assisted generation, manually reviewed and curated by the author
• Drug name source: Sample drawn from a publicly available Indian pharmaceutical database

Samples were generated by prompting GPT-4o to simulate STT autocorrection behaviour for a given Indian medication brand name and produce a naturalistic patient sentence containing the misconstrued phrase. All generated samples were manually reviewed; those that were clinically implausible, linguistically unrealistic, or where the misconstrued phrase was insufficiently distinct from surrounding context were corrected or discarded. Approximately 12% of generated samples were dropped and about 30% were substantially revised.

Schema

Field	Description
true_drug_name	Correct Indian medication brand name
context_conditions	Medical condition(s) the drug is prescribed for
raw_stt_transcript	Full patient sentence as it would appear from STT output
misconstrued_phrase	The specific word or phrase substituted in place of the drug name

Sample

true_drug_name	context_conditions	raw_stt_transcript	misconstrued_phrase
Azithral 500 Tablet	bacterial infection, UTI	My doctor prescribed as it roll yesterday.	as it roll

Limitations

• Single LLM used for generation; hallucination patterns may not fully represent real diverse STT autocorrection behaviour
• Single-drug mention per sentence only
• Generic drug names (paracetamol, ibuprofen, etc.) not included
• English only; no code-mixed sentences

Citation

@dataset{indmedstt_synthv1_2026,
  author    = {Osborne Pereira},
  title     = {IndMedSTT-Synth: Synthetic STT Hallucination Dataset
               for Indian Medication Brand Names},
  year      = {2026},
  publisher = {Project IndRAH},
  url       = {https://huggingface.co/datasets/osbornep8/IndMedSTT-Synth},
  license   = {CC BY 4.0},
  version   = {1.0}
}

License

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

View License →

For questions or collaboration, reach out via Hugging Face Discussions or contact the author at pereiraosborne8@gmail.com.