Scripts & Linguistics

Sandhi

Sandhi is the phonetic junction where adjacent sounds merge or modify each other — critical in Indian languages for correct pronunciation.

VoisLabs TeamUpdated March 2026

Sandhi (Sanskrit: सन्धि, "joining") is a phonological process where adjacent sounds at word or morpheme boundaries modify each other — the Sanskrit term has been adopted into Western linguistics as the general name for this process across all languages. Sandhi appears in most Indian languages (Sanskrit, Tamil, Malayalam, Hindi, Bengali, Marathi) and is typically classified as either external sandhi (applying across word boundaries — "car park" → "carpark" pronunciation in English is a mild example) or internal sandhi (applying within a word at morpheme boundaries). In Sanskrit and Tamil, sandhi is rigorously systematised with specific phonetic rules — a vowel + vowel combination, a consonant + consonant combination, or a consonant + vowel combination each triggers a predictable transformation. Sandhi is one of the features that makes Indian-language TTS difficult: a system that treats each word in isolation will mispronounce content where two words trigger sandhi. Good Indian-language TTS applies sandhi rules during text normalisation, producing natural-sounding speech at word junctions.

How it works

In Sanskrit, sandhi is so systematic that it's essentially a set of algebraic rules: /a/ + /a/ = /aː/ (long a), /a/ + /i/ = /eː/, /i/ + /a/ = /ya/, etc. Students of classical Sanskrit memorise hundreds of sandhi rules. Tamil sandhi includes glide insertion (/i/ + vowel → /i/ + y + vowel), consonant assimilation, and context-dependent voicing changes. Malayalam sandhi is particularly complex because it's explicitly written out — whereas Sanskrit sandhi often appears in the pronounced form only, Malayalam spells out the result of sandhi in the written script, producing the conjunct-heavy appearance that makes Malayalam text rendering hard. Modern colloquial Hindi applies sandhi less rigidly but still respects common patterns — "अधिक + आर्थिक" → "अधिकार्थिक" blurs the word boundary audibly. TTS systems that apply sandhi at text-normalisation time produce smoother, more natural audio compared to systems that pronounce each word as if isolated.

Examples

Sanskrit vowel sandhi

"अति + अन्त" → "अत्यन्त" (ati + anta → atyanta) — the /i/ + /a/ at the junction becomes /y + a/, merging the words into "limitless".

Tamil sandhi

"கதவு" + "அருகே" → "கதவருகே" — the final /u/ of "kathavu" (door) drops before the initial vowel of "aruke" (near).

Malayalam conjunct from sandhi

"സം + കാരം" → "സംസ്കാരം" (samskaaram, "culture") — the nasal and consonant merge into a written conjunct form.

Why this matters for Indian-language TTS

Sandhi handling is a key quality differentiator for Indian-language TTS. Platforms that handle sandhi correctly produce natural-sounding Sanskrit recitation, devotional content, classical poetry, and formal Tamil/Malayalam. VoisLabs voices are trained on Indian-language speech with sandhi features preserved — the voice output correctly merges sound boundaries, which matters for audiobook narration, scripture reading, and educational content where the audio must match how native speakers actually pronounce.

Frequently Asked Questions

Do all Indian languages apply sandhi?
Most do. Sanskrit and Tamil have rigorous sandhi systems. Malayalam applies sandhi extensively and spells out the result in writing. Hindi and Bengali apply sandhi less systematically but still in common patterns. Modern spoken Hindi is the least sandhi-rigid among major Indian languages.
How does TTS handle sandhi?
Quality Indian-language TTS applies sandhi during text normalisation — the preprocessing stage before the neural model generates audio. Platforms without sandhi handling pronounce each word in isolation, producing artificial-sounding word boundaries. Native-speaker listeners detect this immediately.
Why does sandhi matter for karaoke subtitles?
Karaoke subtitles highlight words as they're spoken. If the TTS applies sandhi and merges word boundaries, the highlight timing must match the merged audio — another layer of complexity. VoisLabs' karaoke subtitle engine is sandhi-aware.

Try VoisLabs — Indian-language TTS done right

1 minute free per day. 12 languages. Native Indian-script karaoke subtitles. No card required.

Start free

Last verified: 2026-04-21