Custom Vocabulary for Indian Names & Brands

Indian names come from dozens of languages and have no single standard romanization. 'Shrinivas' might be spelled Srinivas, Shreenivas, or Sreenivasan depending on the family's preference. The AI cannot know which variant a specific person uses - it picks the statistically most common spelling, which is often wrong.

Brand names face similar challenges. Zerodha, Byju's, Cred, Swiggy, Zomato - these are common enough to be in AI vocabularies. But regional brands, startup names, and new companies are not. A fintech startup called 'Khatabook' might be transcribed as 'khata book' or 'kata book'.

Before transcribing, list every proper noun you expect to hear: people's names with their preferred spelling, company and brand names, product names, place names (especially smaller cities or neighborhoods), and any technical jargon specific to the topic.

For recurring projects - a weekly podcast, regular meeting series - maintain a running vocabulary file. Add new terms after each episode and carry the list forward. Over time, this becomes a comprehensive reference that dramatically reduces review time.

When transcribing into Devanagari, names need correct Hindi rendering. 'Priya' should be प्रिया not प्रीया. 'Rahul' should be राहुल not रहुल. These are obvious to native speakers but AI frequently gets matra length wrong on names because the model treats them as any other word.

For names that originate from one language but appear in another's script - a Tamil name in a Hindi transcript, or an English name in Devanagari - decide in advance how to render them. Consistency matters more than any single correct choice.

After AI transcription, search the output for each term on your list. Check that it appears with the correct spelling every time. AI inconsistency means the same name might appear three different ways in one transcript. Use find-and-replace to standardize.

Also search for phonetic variants the AI might have produced. If the name is 'Subramanian', search for 'Subra' to catch 'Subramaniam', 'Subramanyam', or 'Subramani'. For Hindi names, search for the first syllable in Devanagari to catch matra variants.

Legal: Indian legal terms mix Hindi, English, and Urdu. FIR (First Information Report), vakeel, adalat, rapat - these have standard spellings that AI may not know. Medical: Ayurvedic terms, Hindi medical vocabulary, and drug names need explicit inclusion. Tech: Indian startup ecosystem jargon (Series A, ARR, DAU) mixes with Hindi business terms.

For each domain, spend 5-10 minutes listing terms before the first transcription. This one-time investment pays off across all future recordings in that domain.